MSVC complains about an implicit conversion from double to int.
Also small comments tweaks.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
If null search fails high return null value instead of beta.
With TT hash there may be a small advantage for fail-soft since
storing slightly better bounds may cause slightly more hash hits.
After 990 games at 1+0
Mod vs Orig +171 =665 -154 +6 ELO
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Array castleRightsMask[] is not static because it can
be different for different positions, so let it be
a Position member data. This allows to remove tricky
hacks to take in account that although it was defined
static it could change.
Theoretically now copying a position is a bit slower because
we need to copy also an array of 64 integers, but because in
split() we don't copy the position anymore, but just keep the
pointer, the added burden is not mesurable even in MP case.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Also don't use __cpuid() intrinsic for Intel under
Linux because gives wrong results when detecting HT,
use the gcc version instead. Finally clean up the code.
Error was due to changed __cpuid() signature for
gcc compiler.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
This will allow to use the function also for other
purposes then detecting POPCNT.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It will be used by future patches and also rearranges some
half cooked code that mistakenly ended up in master in the
past.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Currently, in case of fail high/low we research with
a window increased by 2*AspirationDelta at first
attempt, this patch instead makes the research be
done with an increase of just AspirationDelta size,
in case of a consecutive fail we will widen to
2*AspirationDelta and so on.
After Joona's test:
Orig - Mod: 850 - 890
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
One for failing highs and one for failing lows, this
should reduce average window size in case of positions
that fail first high and then low (or the contrary).
After ~2000 games on Joona's quad we have:
Mod - Orig: 1012- 975 (+6 elo)
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Instead to leave uninitialized or scattered in the code
as is the case for ExtraSearchTime.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Currently we use original sorting after a fail low to
research at wider window. This patch instead sorts the
moves according to the last available move's scores.
Strangely no functional change, but should be.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Changed FutilityMarginsMatrix dimensions to be a power of two
so that compiler can produce a faster accessing code.
Introduced print_pv_info() to remove some redundant code in
root_search()
Remaining stuff is triviality and documentation tweaks.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Under Windows we use CreateThread() to setup threads and
we pass a pointer to a variable that receives the thread
identifier, but this parameter is optional and we don't
use it, so remove it.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It happens that NULL is 0, but the conventional meaning is of
a zero pointer, so repleace with an explicit 0 integer value.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Consider sligtly negative captures as good if at low
depth and far from beta.
After 999 games at 1+0
Mod vs Orig +169 =694 -136 +11 ELO
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It raises an assert under Windows, it is not clear why but it
happens that idle_loop() is called with incorrect threadID and
the assert triggered is:
assert(threadID >= 0 && threadID < MAX_THREADS);
So revert the patch for now, but we should understand why it
fails.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
We can't do it with full guarantee anyway because
there is always a possible race between the setting of
state to THREAD_SLEEPING and actual sleeping.
So just remove the not perfect code to avoid misunderstandings.
This reflects what we have done in wake_sleeping_threads() in
the previous patch.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Currently there is no guarantee that threads are sleeping
when calling wake_sleeping_threads() because put_threads_to_sleep()
returns without waiting for threads to actually sleep.
Assert can be easily triggered calling put_threads_to_sleep() and
wake_sleeping_threads() in a tight loop.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
So we can use a const value instead of a pointer in
split().
Also pass NULL instead of a faked address of alpha in
case split is called from a non-PV node.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Intrinsic __popcnt64() returns an unsigned __int64, cast
to an integer and silence the warning.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Joona says that sp_update_pv() does not pass the split point
boundaries, so there is no risk to corrupt data from another
split point. Also the race on thread_should_stop() is harmless
because of this.
So revert the patch and come back to single lock.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Joona says:
1. We should not be afraid of "AllThreadsShouldExit" flag.
Because when this is set to true we _must_not_ be searching (= All
splits must have been undone).
And if we are not searching it's impossible that some other thread
could give us work to do. So setting state to THREAD_AVAILABLE
doesn't do any harm. If you want to add check for this, you could do
it like this:
if (threads[threadID].state == THREAD_WORKISWAITING)
{
+ assert(!AllThreadsShouldExit)
threads[threadID].state = THREAD_SEARCHING;
2a. If waitSp->cpus == 0, setting state to THREAD_AVAILABLE makes
no harm either, because helpful master concept dictates that _only_
our own slave can book us. If we don't have any slaves, noone has the
right to book us.
2b. If point (2a) is not correct then your extra check only adds extra race:
In smp code checking for waitSp->cpus > 0 is not enough. It's possible that
our slave immediately exits and another thread
books us as a slave when our state is still
THREAD_AVAILABLE. So instead of adding extra level of security we have
just introduced extra race.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
When we found a cut-off then lock all the split point chain,
not only current one to avoid races in case two threads running
on different split points where one is ancestor then the other,
find a beta cut-off at the same time, in this case we want only
one to call sp_update_pv().
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
This is a per split-point request, not per-thread. When we find
a beta cut-off in current thread's split point or in or in some
ancestor of the current split point then threads should stop
immediately the search and return to idle_loop().
The check is done by thread_should_stop() that now looks only
at split point's chain.
No functional change and a good semplification.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
This is easier to follow and also reduces the points
where state changes to mainly idle_loop() and split().
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
In this case is dangerous because in split() we reset the flag to
false, but if it was set due to a cut-off higher in the tree we
completely miss that and go on with the full search.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Instead of other flags this is not a state flag, i.e. does
not defines a state for the thread, but a request because
after we raise 'stopRequest' flag the corresponding thread is
not stopped, but continues to run for a while until it returns
from sp_search() in idle_loop.
It is important the name reflects this.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Among them 'stop' and 'printCurrentLineRequest' could have
random value, so reset to a known state before to leave the
search.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
This is the first nice effect of previous patch !
Because thread_should_stop() should be declared 'const' we
need to remove the setting of 'stop' flag to true that
turns out to be a bug because thread_should_stop() is called
outside from lock protection while 'stop' flag is a volatile
shared variable so cannot be changed when not in lock.
Note that this bugs fires ONLY when we use more then 2 threads,
so commonly only in a QUAD or OCTAL machine.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Main aim of this patch is to consolidate all the thread related stuff
behind a single class interface so to avoid messing with global flags
and having thread code scattered among non-thread related stuff.
Another advantage is that now access to thread's variables is
more controlled, in particular we can differentiate between
read and write accesses by the mean of different interfaces, it
is so simpler to understand how a function is related to threads.
Lastly this rewrite is the base for future code consolidations and
semplifications that are easier now that we have only one thread's
access point.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Ensure threads are sleeping when leaving init_threads() and
the newly introduced put_threads_to_sleep().
Also ensure threads are not sleeping when leaving
wake_sleeping_threads().
As a side effect we now leave think() with all the threads
(but the main one) guaranteed to sleep. So when we enter
again in think(), after the opponent next move, we know
threads must be sleeping.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Wait inside wake_sleeping_threads() for the threads to be
effectively and reliably woken up.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Will be used by future patches. Also:
- Renamed Idle in AllThreadsShouldSleep
- Explicitly inited AllThreadsShouldExit and AllThreadsShouldSleep
in init_thread() instead of use an anonymous global initialization.
- Rewritten idle_loop() while condition to avoid a 'break' statement
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Patch from Richard Lloyd (slightly edited from me), following the list
of changes as described by the author:
src/Makefile:
- Added PREFIX and BINDIR for the install: rule.
- Added a "make hpux" line to the help: rule.
- Added "make test"/"make check" rule that runs the $(PGOBENCH) command.
- "make clean" now additionally removes core and bench.txt.
- Added an hpux: rule.
- Added an install: rule to mkdir $(BINDIR), copy $(EXE) to $(BINDIR) and
then strip it.
- "make strip" now ensures that $(EXE) is built first before trying to
strip it.
- Hide errors and output from the g++ command used by the .depend: rule and
then touch .depend in case g++ isn't available.
- Hide errors from the "include .depend" in case .depend doesn't exist
(e.g. directly after a "make clean").
src/book.cpp and src/book.h:
- HP-UX's aCC really didn't like the const keywords used for the
Book::file_name() definitions, so they were removed. I checked that this
didn't affect a Linux build and it was still fine.
src/misc.cpp:
- HP-UX uses <sys/pstat.h> and pstat_getdynamic() to determine the number of
CPU cores, so added conditional code for that (if pstat_getdynamic() fails,
set the number of cores to 1).
src/tt.cpp:
- <xmmintrin.h> and _mm_prefetch() seem highly specific to the Intel x86(_64)
and gcc platforms - neither exist in HP-UX, so conditionally avoid that
code in HP-UX's case. Perhaps some sort of define is needed here
such as -DHAS_MM_PREFETCH that could be #ifdef'ed for instead?
Even after these changes, it's more convenient for HP-UX users to edit the
default: rule in the Makefile to run "$(MAKE) hpux" before they build
stockfish, but that's not a big deal if they're warned about that first (the
same applies to all other builds other than the standard "$(MAKE) gcc" one).
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Compiler complains because in Book we have a d'tor but not
copy c'tor and assignement operator (warning C4511 and C4512),
note that after adding them (just declared) you now need also
default c'tor !
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It is an hidden bug waiting to fire. The main problem is
that ss[ply] is overwritten by search() and qsearch() called
from IID and razoring, so that we cannot hold a pointer to a
local EvalInfo variable.
For instance if we go razoring then we overwrite the pointer
with the address of a variable local to qsearch(), when we return
from qsearch() variable goes out of scope and now ss[ply].evalInfo
holds a stale pointer !
Because we are not looking for troubles we go through the
safe route and we remove it entirely.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Gain value is multiplied by 16 to be of comparable magnitudo
of negative history, on average.
This patch shows very good results in tactical tests, but
started very bad in real games, so I have run two test matches.
After 896 games at 1+0
Mod vs Orig +187 =525 -184 +1 ELO
After 999 games at 1+0
Mod vs Orig +223 =590 -186 +13 ELO
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
We can do this only when needed, if we get a cut-off
before we skip sorting entirely. This reduces sorting
time of about 20%.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
With negative history we don't have anymore a
lot of zeroes to score, so just split moves in
positives and non-positives sets.
Speed up is almost zero, we cannot test speed directly
because node count changed due to reorder, but I have
verified sorting is correct. With a profiler I have
seen we gain a little in sort_moves() and lose a little
in insertion_sort(), so the net effect is almost zero,
but code is simpler.
No real change, just move reordering.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Because H.move_ordering_score() can return negative values
some negative see moves could be searched before non-negative
see moves with negative history.
This patch restores proper ordering.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Now that history can go negative and is almost alwyas
non zero we have no more reasons to use also psqt term.
After 994 games at 1+0
Mod vs Orig +204 =597 -193 +4 ELO
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Instead of piece-from-to, in this way it is similar
to what we already do for history.
Almost no change, but seems a bit simpler in this way.
After 995 games at 1+0
Mod vs Orig +207 =596 -192 +5 ELO
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It seems more standard conformant. Also added a bit of
description directly from Tord.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
We erroneusly added two times the same scaling function
to endgame's map.
Fix detected by valgrind becasue resulted in a memleak
of the first added scaling function.
Bug introduced by 30e8f0c9ad6a473 of 13/02/2009
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
We had an overflow due to use an integer for hash size,
now we use a size_t as we should, so we can increase to
an higher limit.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Normally it's up to the GUI to check for option's limits,
but we could receive the new value directly from the user
by teminal window. So let's check the bounds anyway.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
With new target 'make gcc-popcnt' it is now
possible to compile with enabled hardware POPCNT
support also with gcc. Until now was possible only
for Intel and MSVC compilers.
When this instruction is supported by CPU, for instance
on Intel i7 or i5 family, produced binary is a bit faster.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
This reverts commit c43c5fe9e0.
Produces following build error after 'make clean', 'make icc' under Mandriva
with icc version 11.0
Makefile:306: .depend: No such file or directory
In file included from tt.cpp:28:
/usr/lib/gcc/i586-manbo-linux-gnu/4.3.2/include/xmmintrin.h:35:3: error: #error "SSE instruction set not enabled"
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Futility captures alone does not seem an improvment.
Perhaps is a combination of stand pat + futility that is winning,
so revert for now and continue testing starting from a standard
base until we find the correct receipe.
After 999 games at 1+0
Mod vs Orig +231 0506 -201 +10 ELO
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Under some rare cases we can have a search tree explosion
due to a perpetual check or to a very long non-capture TT
sequence.
This avoids the tree explosion not following TT moves that
are not captures or promotions when we are below the
'generate checks' depth.
Idea suggested by Richard Vida.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Not a biggie but is a reduced pruned patch that doesn't
seems to hurt, so it is welcomed ;-)
After 999 games at 1+0
Mod vs Orig +207 =601 -191 +6 ELO
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
At ply == OnePly (common case) we avoid some useless
floating point computation.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Joona new aspiration window. Main idea is to always
research aspiration fail highs/low at the same
ply and use much smaller aspiration window than previously.
Testing result is very positive.
1CPU:
953-1149
4CPU:
545 - 656
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Avoid to take the lock two times in a tight sequence, the first
in get_next_move() and the second to update sp->moves.
Do all with one lock and so retire the now useless locked version
of get_next_move().
Also fix some theorical race due to comparison sp->bestValue < sp->beta
is done out of lock protection. Finally fix another (harmless but time
waster) race that coudl occur because thread_should_stop() is also
called outside of lock protection.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
In debug run with 2 threads it happens to be following
assert after some minutes:
assert(value > -VALUE_INFINITE && value < VALUE_INFINITE);
in search(), line 1615.
I am not able to understand why, anyhow reverted for the moment.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
This avoid us to forget some very needed tests now that
futility has changed in a whole big chunk we need to fine
tuning every splitted change.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
This will be useful to use gains table in move
ordering along with history table.
No functional change and big code remove.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
In less then 1% of cases value > sp->bestValue, so avoid
an useless lock in the common case. This is the same change
already applied to sp_search().
Also SplitPoint futilityValue is not volatile because
never changes after has been assigned in split()
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
When we have more then 2 threads then we do an array
access with index 'Threads[slave].activeSplitPoints - 1'
This should be >= 0 because we tested the variable just
few statements before, but because is a shared variable
it could be that the 'slave' thread set the value to zero
just after we test it, so that when we use the decremented
variable for array access we crash.
Bug spotted by Bruno Causse.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Unify start loop for master and slave threads. Also guarantee
that all the 'stop' flags are set to false before first slave
is started, should be no harm because only master thread can
reset 'stop' flag of slaves to true, so should be no race but
better safe then sorry.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Only pure blocking evasions are candidate
for pruning.
After 998 games at 1+0
Mod vs Orig +215 =596 -187 +10 ELO
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
In move_to_san() we create by copy a new position just
to detect if move gives check. This could be very costly in
line_to_san() that calls move_to_san() for every move, so
create the position only once and pass a reference to move_to_san()
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
When pondering threads are put to sleep, but when thinking
the threads are parked in idle_loop in a tight polling loop
checking for workIsWaiting falg.
So before we set the slave's flag workIsWaiting we have to
guarantee that all the slave data is already setup because
slave can start in any moment from there.
Rearrange the last loop to fix this race.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Once we have allocated our slave threads and we have removed
master from available threads we can safely remove the lock
so that the lenghty search stack copy operation will not
impact lock contention.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
A pointer is enough because after a split point has been
setup master and slaves thread end up calling sp_search() or
sp_search_pv() and here a full copy of split point position is
done again, note that even master does another copy (of itself)
and this is done before any do_move() call so that master Position
is never updated between split() and sp_search().
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
And detach splitPoint Position from the master one.
So we duplicate StateInfo only once in split() instead
of one for each thread in sp_search
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Renamed a bit the functions to be more clear what
we actually are doing when we craete a Position object
and explained how StateInfo works.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Also removed some trailing whitespaces and aligned
indentation to current standard.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Only the previous, the current and the next ply SearchStack
are copied.
This reduces split overhead especially at low depth (high ply)
and with many threads.
Possibly no functional change (it is not easy to prove in SMP)
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It is a good programming practice to verify a system
call has indeed succeed.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
When a search fails high then sp->alpha is increased and
slave threads are requested to stop.
So we have to check for a stop request before to start a search
otherwise we could end up with sp->alpha >= sp->beta
leading to an assert in debug run in search_pv().
This patch fixes the assert and get rid of some of possible races.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
In qsearch() try to get a cutoff with the help of an
extra check if we are already very near.
Small increase in actual games but a good result in tactical
test sets where this patch makes SF more tactical.
Mod vs Orig +197 =620 -181 +6 ELO
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Louis Zulli reports a miscompile with g++-4.4 from MacPorts.
Namely enum Value is compiled as unsigned instead of signed integer
and this yields an issue in score_string() where float(v) is incorrectly
casted when Value v is negative.
This patch ensure that compiler choses a signed variable to store a Value.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Value is never used un-initialized, but MSVC is not
smart enough to detect itself :-(
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Marco Costalba