If we don't have tt moves to search skip the
useless loop associated with TT_MOVES phase.
Another 1% speed boost that brings this series
to a +6.2% against original revision 595a90df
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
This not only cleans up the code but gives another
speed boost of 1.8%
From revision 595a90dfd0 we have increased pgo compiled binary
speed of a whopping +5.2% without any functional change !!
This is really awsome considering that we have also
cut line count by 25 lines.
Sometime we spend days for getting an extra 1% from move
generation while instead the biggest optimizations come
from anonymous and apparently dull parts of the code.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Does not seem to improve on the standard, latest results
from Joona after 2040 games are negative:
Orig - Mod: 454 - 424 - 1162
And is more or less the same I got few days ago.
So revert for now.
Verified same functionality of 595a90dfd
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Use the same way of loop along the move list used for
the others move kinds so to be consistent in get_next_move()
And a bit of the usual clean up too, but just a bit.
It is even a bit (+0.3%) faster now. ;-)
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Always after TT move but before captures.
This seems a better setup against version before this
patch.
After 999 games at 1+0
Mod - Orig +252 =527 -220 +11 ELO
Unfortunatly it does not seems to improve on the standard
version, with null move outside of movepicker (595a90df) with
the latest speed-up patches added in.
After 999 games at 1+0
Mod - Standard +244 =506 -249 -2 ELO
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
This avoids calculating the array entry position
at each access and gives another boost of almost 1%.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
In MovePicker we get the next move with pick_move_from_list(),
then check if the return value is equal to MOVE_NONE and
in this case we update the state to the new phase.
This patch reorders the flow so that now from pick_move_from_list()
renamed get_next_move() we directly call go_next_phase() to
generate and sort the next bunch of moves when there are no more
move to try. This avoids to always check for pick_move_from_list()
returned value and the flow is more linear and natural.
Also use a local variable instead of a pointer dereferencing in a
time critical switch statement in get_next_move()
With this patch alone we have an incredible speed up of 3.2% !!!
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Try good captures before null move when depth < 3 * OnePly.
Use this kind of null move also in Depth == OnePly.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Original patch from Joona with added optimizations
by me.
Great cleanup of MovePicker with speed improvment of 1%
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
In Movepicker c'tor we access during initialization one of
MainSearchPhaseIndex..QsearchWithoutChecksPhaseIndex globals.
Postpone definition of PhaseTable[] just after them so that
when PhaseTable[] will be accessed later in get_next_move()
it will be already present in L1/L2.
It works like an implicit prefetching of PhaseTable[].
Also shrink PhaseTable[] to fit an L1 cache line of 16 bytes
using uint8_t instead of int.
This apparentely innocuous patch gives an astonish speed
up of 1.6% under MSVC 2010 beta, pgo optimized !
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
When ordering moves we push all captures with negative SEE values
to badCaptures[] array during the scoring phase.
This patch delays the costly SEE call up to when the move has been
picked up in pick_move_from_list(), this way we save some SEE calls
in case we get a cutoff.
It seems we have a speed gain of about 1-1.5 % in terms of nodes/sec
and profiling seems to confirm the small but real speed increase.
Idea from Pablo Vazquez on talkchess.com
http://www.talkchess.com/forum/viewtopic.php?t=29018&start=20
It would be a no functional change but actually it is not because
now sorting set is different and so std::sort(), that is not a
stable sort, does not guarantees the order of same scored moves to
remain the same as before.
After 952 games at 1+0 we are below error bar, almost equal just
6 games of difference (+2 ELO)
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Mostly of times we are interested only in the sign of SEE,
namely if a capture is negative or not.
If the capturing piece is smaller then the captured one we
already know SEE cannot be negative and this information
is enough most of the times. And of course it is much
faster to detect then a full SEE.
Note that in case see_sign() is negative then the returned
value is exactly the see() value, this is very important,
especially for ordering capturing moves.
With this patch the calls to the costly see() are reduced
of almost 30%.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Instead of add and subtract pqst values corrisponding to
the move starting and destination squares, do it in one
go with the helper function pst_delta<>()
This simplifies the code and also better documents that what
we need is a delta value, not an absolute one.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Rename to move_is_promotion() to be more clear, also add
a new function move_promotion_piece() to get the
promotion piece type in the few places where is needed.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
This was needed by an old optimization in sorting of
non-captures that is now obsoleted by new std::sort()
approach.
Remove also the unused depth member data. Interestingly
this has always been unused since the Glaurung days.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
This is a first step for future patches and in
any case seems a nice thing to do.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
In MovePicker consider killer moves as a separate
phase from non-capture picking.
Note that this change guarantees that killer1 is always
tried before killer2. Until now, because scoring difference
of the two moves was just 1 point, if psqt tables of killer1
gave a lower value then killer2, the latter was tried as first.
After 999 games at 1+0 we have
Mod vs Orig: +245 =527 -227 +6 ELO
Not a lot but patch is anyhow something worth to have.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Both with added comment and changing the API to
reflect that only destination square and moved piece
is important for history.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Instead of a delayed selection sort so that the highest
score move is picked up from the list when needed, sort all
the moves up front just after score them.
Selection sort is O(n*n) while std::sort is O(n*log n), it
is true that delayed selection allows us to just pick the move
until a cut off occurs or up to a given limit (12), but with
an average of 30 non capture-moves delayed pick become slower
just after 5-6 moves and we now pick up to 12.
Profiling seem to prove this idea and movepick.cpp is now 10%
faster.
Also tests seem to confirm this:
After 700 games at 1+0: Mod vs Orig +178 -160 =362 +9 ELO
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Instead of loop across all legal moves to find a mate
loop across possible check moves only.
This reduces more then 10 times the number of moves to
check for a possible mate.
Also rename generate_checks() in generate_non_capture_checks()
so to better clarify what the function does.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
And upon reentering the same position try it as first.
Normally qsearch moves order is already very good, first move
is the cut off in almost 90% of cases. With this patch, we get
a cut off on TT move of 98%.
Another good side effect is that we don't generate captures
and/or checks when we already have a TT move.
Unfortunatly we found a TT move only in 1% of cases. So real
impact of this patch is relatively low.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Instead of just drop evaluation score after stand pat
logic save it in TT so to be reused if the same position
occurs again.
Note that we NEVER use the cached value apart to avoid an
evaluation call, in particulary we never return to caller
after a succesful tt hit.
To accomodate this a new value type VALUE_TYPE_EVAL has been
introduced so that ok_to_use_TT() always returns false.
With this patch we cut about 15% of total evaluation calls.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Reduce of two plies near the leafs and when we still
have enough depth to go so to limit horizon effects.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It is slower the previous uglier but faster code.
So completely restore old one for now :-(
Just leave in the rework of status backup/restore in do_move().
We will cherry pick bits of previous work once we are sure
we have fixed the performance regression.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
It happens that more then 70% of cases are HxL, where
we call see() anyway. The mesured saving of calling
see is about 0,5% of total time, but considering the
added burden in score_captures() the saving is only
0,35% locally and due to more difficult inlining of
the function it ends up that we have no advantage at all,
possibly a small slow down!
So revert.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Does not seem to improve anything.
Anyhow idea is nice, maybe we still have to find
correct recipe.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
EvalInfo has missing attack info when a specialized
endgame function is used.
We missed this case and were using an empty attack bitboard
instead so that no captures were generated for endgames.
After testing the EvalInfo optimization gave worst results,
so after a (long) debug session this nasty bug was found.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Add infrastructure to threat killer moves as a vector,
this will allow us to easily parametrize the number of
killer moves, instead of hardcode this value to two as is now.
This patch just add the infrastructure, no functional
change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
