Hex representation doesn't add any value in those cases.
Preserve hex representation where more self-documenting
for instance for binary masks values.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It seems to increase SMP performances. To note that
this patch goes in the opposite direction of "Active
reparenting" where we try to reparent an idle slave
as soon as possible. Instead here we prefer to keep
it idle instead of splitting on a shallow / near the
leaves node.
After 11550 games on a QUAD (4 threads) at 15"+0.05
Mod vs Orig 1972 - 1752 - 7826 ELO +6 (+-3.6)
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Set optimization level to 4 and get a 2.564% faster binary:
Stockfish (Clang, Level 4) bench:
$ make build ARCH=osx-x86-64 COMP=clang
(Clang does not support PGO)
Average of 4 trials:
Total time (ms): 5137.5
Nodes searched: 5631135
Nodes/second: 1096084.5
Stockfish (Clang, Level 3) bench:
$ make build ARCH=osx-x86-64 COMP=clang
(Clang does not support PGO)
Average of 4 trials:
Total time (ms): 5269.25
Nodes searched: 5631135
Nodes/second: 1068679.25
Stockfish (GCC, PGO) bench:
$ make profile-build ARCH=osx-x86-64
Average of 4 trials:
Total time (ms): 5286
Nodes searched: 5631135
Nodes/second: 1065292.25
Suggestion and performance tests by Daylen Yang.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
A PvNode that givesCheck has been already granted
an extension of ONE_PLY in previous condition.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Makefile modified to support the clang compiler under Mac.
This was tested using clang 4 under Mountain Lion, but should
also work fine under Lion and possibly under Snow Leopard.
It requires the 'Xcode 4.x Command Line Tools' to be installed.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Since revision 374c9e6b63
we use also castling information to calculate king safety.
So before to reuse the cached king safety score we have to
veify not only king position, but also castling rights are
the same of the pre-calculated ones.
This is a very subtle bug, found only becuase even after
previous patch, consecutives runs of 'bench' _still_ showed
different numbers. Pawn tables are not cleared between 'bench'
runs and in the second run a king safety score, previously
evaluated under some castling rights, was reused by another
position with different castling rights instead of being
recalculated from scratch.
Bug spotted and tracked down by Balint Pfliegel
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Now that we can call bench multiple times
from command prompt we need to ensure searched
nodes remain constant across different runs.
Spotted by Bálint Pfliegel.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
After 6K games at 60" + 0.1 on QUAD with 4 threads
this implementation fails to show a measurable increase,
result is well within error bar.
Perhaps with 8 or more threads resut is better but we
don't have the hardware to test. So retire for now and
in case re-add in the future if it proves good on big
machines.
The only good news is that we don't have a regression and
implementation is stable and bug-free, so could be reused
somewhere in the future.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
The check for detecting when a split point has all the
slaves still running is done with:
slavesMask == allSlavesMask
When a thread reparents, slavesMask is increased, then, if
the same thread finishes, because there are no more moves,
slavesMask returns to original value and the above condition
returns to be true. So that the just finished thread immediately
reparents again with the same split point, then starts and
then immediately exits in a tight loop that ends only when a
second slave finishes, so that slavesMask decrements and the
condition becomes false. This gives a spurious and anomaly
high number of faked reparents.
With this patch, that rewrites the logic to avoid this pitfall,
the reparenting success rate drops to a more realistical 5-10%
for 4 threads case.
As a side effect note that now there is no more the limit of
maxThreadsPerSplitPoint when reparenting.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Check for a cutoff occurred also high in
the tree and not only at current split
point.
This avoids some more wasted reparenting.
No functional chnage.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It is more correct given what the function does. In
particular single_bit() returns true also in case of
empty bitboards.
Of course also the usual renaming while there :-)
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Instead of reparenting to oldest split point, try to reparent
to latest. The nice thing here is that we can use the YBWC
helpful master condition to allow the reparenting of a split
point master as long as is reparented to one of its slaves.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
And update master->splitPointsCnt under lock
protection. Not stricly necessary because
single_bit() condition takes care of false
positives anyhow, but it is a bit tricky and
moving under lock is the most natural thing
to do to avoid races with "reparenting".
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
In Young Brothers Wait Concept (YBWC) available slaves are
booked by the split point master, then start to search below
the assigned split point and, once finished, return in idle
state waiting to be booked by another master.
This patch introduces "Active Reparenting" so that when a
slave finishes its job on the assigned split point, instead
of passively waiting to be booked, searches a suitable active
split point and reprents itselfs to that split point. Then
immediately starts to search below the split point in exactly
the same way of the others split point's slaves. This reduces
to zero the time waiting in idle loop and should increase
scalability especially whit many (8 or more) cores.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Apart from the semplification it is now more clear that
the actual Tempo added was half of the indicated score.
This is becuase at start compute_psq_score() added half
Tempo unit and in do_move() white/black were respectively
adding/subtracting one Tempo unit.
Now we have directly halved Tempo constant and everything
is more clear.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It is still enabled during fixed limit search so to
use it during fixed depth/nodes/time matches.
Bug reported by Daylen.
No functional changes.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Shrinking from [16] to [2][2] is able to speedup
perft of start position of almost 5% !
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Shrink dimensions of the biggest stack consumers arrays.
In particular movesSearched[] can be safely shrinked
without any impact on strenght or risk of crashing.
Also MAX_PLY can be reverted to 100 with almost no impact
so to limit search recursion and hence stack allocation.
A different case is for MAX_MOVES (used by Movepicker's
moves[]), because we know that do exsist some artificial
position with about 220 legal moves, so in those cases SF
will crash. Anyhow these cases are never found in games.
An open risk remains perft, especially run above handcrafted
positions.
This patch originates from a report by Daylen that found
SF crashing on his Mac OS X 10.7.3 while in deep analysys
on the following position:
8/3Q1pk1/5p2/4r3/5K2/8/8/8 w - - 0 1
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Now a fen file with Chess960 positions is
correctly parsed. But it is mandatory to set
"UCI_Chess960" option _before_ to call bench.
Note that this was not needed/possible before
adding the possibility to call 'bench' from
command prompt.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Now that we can call bench on current position
we can directly use it to perform our perft.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Now that we can call bench from command prompt
has a sense to teach bench to run the current
set position. To do this is enough to call bench
with 'current' as fen source parameter.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
After extensive test Gary says:
"So, after 16k games at 10"+1" on an i7, the undefended rook test
looks to be not good (albeit by a very small margin).
3063 - 3093 - 9844 (-1).
I doubt that is causing the regression, but even so, it looks like
it's not worth keeping, and we can go back to the simpler undefended
minors check."
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Unfortunatly accessing thread local variable
is much slower than object data (see previous
patch log msg), so we have to revert to old code
to avoid speed regression.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Much faster then pthread_getspecific() but still a
speed regression against the original code.
Following are the nps on a bench:
Position
454165
454838
455433
tls
441046
442767
442767
ms (Win)
450521
447510
451105
ms (pthread)
422115
422115
424276
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
After we release the SplitPoint lock the master, suppose
is main thread, can safely return and if a "quit" command
is pending, main thread exits and associated Thread object
is freed. So when we access master->is_searching a crash
occurs.
I have never found such a race that is of course very rare
becuase assumes that from lock releasing we go to sleep for
a time long enough for the main thread to end the search and
return. But you can never know, and anyhow a race is a race.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
So to avoid a crash when setting the moves in
UCI "position startpos moves ...." command.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
But use the newly introduced local storage
for this. A good code semplification and also
the correct way to go.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Use thread local storage to store a pointer to the thread we
are running on. This will allow to remove thread info from
Position class.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
With this change sources are fully endianess
independent, so we can simplify the Makefile.
Somewhat surprisingly we don't have any speed
regression !
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Erroneusly adds default positions to the fens
loaded from external file.
Bug introduced in adb71b8096
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
The 2 overload functions map() accept a pointer to
EndgameBase<Value> or a pointer to EndgameBase<ScaleFactor>.
Because Endgame<E> is derived from one of them we can
directly use a pointer to this class to resolve the
overload as is needed in Endgames::add().
Also made class Endgames fully parametrized and no more
hardcoded to the types (Value or ScaleFactor) of endgames
stored.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
A std::set (that is a rb_tree) seems really
overkill to store at most a handful of moves
and nothing in the common case.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It is possible to start with 'stockfish', then from
command prompt type 'bench' and SF will do what you expect.
Old behaviour is anyhow preserved. As a bonus we can now
start from command line any UCI command understood by
Stockfish. The difference is that after execution of a
command from arguments SF quits, while at the end of the
same command from prompt SF stays in UCI loop.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Allows some code semplification and avoids directly
allocation and managing heap memory.
Also the usual renaming while there.
No functional change and no speed regression.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Marco Costalba