MovePicker: introduce per square MVV/LVA ordering

Just added the infrastructure, no functional change.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix hashfull info

Do not count has a replacement when a TT entry is
written in an empty slot.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix in ok_to_history(): castle move is not a capture

It is erroneusly considered a capture because king
moves on the same square of the rook.

Use the correct function Position::move_is_capture()
instead of the open coded (and buggy) one.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Small tidyup of TranspositionTable::store()

Hopefully without bugs this time!

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix a logic bug in TranspositionTable::store()

Make the logic work as advertised in the function
description.

Still a fallback from TT cleanup.

This should be less serious then the one in retrieve(),
but it's still a bug.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Micro optimization of update_history()

Remove an useless comparison.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Use cut-off checks in qsearch as killer moves

Killers should not be captures, but checks are not
and are produced also in qsearch.

Use this information, will be useful for move ordering.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Smaller null move reduction when depth is high

Lower probability to miss something important.

It seems to increase strenght. Idea form Cyclone.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Introduce LastIterations variable

Is set during the last iteration.

Sometime also during the second last.

During the last iteration is set in the 95% of cases.

During the second last is set in the 40% of cases.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Restore development versioning and LSN filtering

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix a missed initialization in get_option_value()

Spotted and reported by Dann Corbit.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Stockfish 1.01

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix compile under Ubuntu 64bit

Some missing includes.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix a serious bug in TranspositionTable::retrieve()

Reported by Tord Romstad.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Restore development versioning

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Restore LSN filtering

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Revert movepick optimizations before to release

More testing is needed and better do not risk
just before release.

Reverted:

Disable LSN filtering as defualt for release
Use MVV/LVA in score_evasions()

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Stockfish 1.0

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Disable LSN filtering as defualt for release

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Use MVV/LVA in score_evasions()

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Delay SEE for scoring captures

Do not calculate SEE on all the moves in MovePicker::score_captures()
but delay until pick_move_from_list() when only the best ones are
double checked against their see value.

If a beta cut-off occurs then we avoid calculating SEE on all
the moves, but just the picked ones.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

search: micro optimization

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix a couple of gcc warnings in position.cpp

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Update README to Stockfish

Remove Glaurung references.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Final touches to material.cpp

No functional changes, altough a bit of code reshuffle.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Material: lockless per-thread maps

Adds a good bunch of code but should be faster
and scalable because is lockless.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Revert "MovePicker::score_captures() order with SEE when pv"

Does not seem to increase the strenght.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Space inflate move.cpp

Also a little cleanup.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Revert "Relax time constraints"

Does not seem to give an improvment, acutually it seems
a totally neutral change.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Partially space inflate search.cpp

Space inflate main remaining functions in search.

No functional change.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Relax time constraints

Allow a seacrh to take a bit more time if needed.

This reduces the chanches of wast all the search time
for the last iteration and also allow to start the last
iteration when we have less time remaining.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Material: micro optimize map reading

Do only one map walk per read instead of two.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Material: protect global map access with locks

STL library is not guaranteed to be thread safe, even for
read-access.

So because these global maps are accessed by all the threads
we need to protect them.

This fixes a random crash experienced during testing.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

MovePicker::score_captures() order with SEE when pv

Order PV nodes by SEE instead of MVV/LVA.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Revert "Last minute surprise" for now

It defenitly needs more testing. Just postponed.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Finish material.cpp cleanup

Hopefully no regression this time!

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Start to simplify material.cpp

It is posisble to simplify a lot here!

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Factor out Position::do_capture_move()

Start to slim line count i position.cpp

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix a bug in generate_evasions()

Introduced in the patch "movegen: prefer (*mlist++) to mlist[n++]"

This was nasty because due to a mismerge the repo in one PC had the bug,
but the testing one did not, so I had non reproducible results according
to which machine I used for testing.

This hopefully closes a more then one week regression that made me go crazy!

It was found by accident comparing, for other reasons, the sources of the
two PC's.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Refine "few moves" in "last minute surprise"

It seems that "few moves" works because we extend the good
captures at the last ply of PV, so code it directly.

This version seems defenitly stronger then previous one.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Manual merge

Space inflate extension() code

No functional change.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

TEST: extend when few moves available

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Revert see shortcut when only one attacker

It does not seem to work after a little testing.

Perhaps it works on the long terms, but it is also
ugly because not correct, so revert for now.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Prefer out of loop variables in MovePicker::score_captures()

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Effectively use MVV/LVA in MovePicker::score_captures()

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Reintroduce piece/square tables to score non-captures

Was removed after original movepick restore. But proved
to be useful.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Position: fix a couple of Intel compiler warnings

Plus usual trailing whitespace.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Test with SEE shortcut

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Better naming of pseudo-legality and legality testing

No functional change.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Try to cleanup movepick

Hopefully without regressions.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Restore original movepick modulo space inflation

We have a regression somewhere here so restart from zero
and proceed one change at a time.

With this modification we have the same strenght of
"Introduce Stockfish" patch that is our strongest to date.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Space inflate position: complete!

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Space inflate position until do_promotion_move()

We will end some day ;-)

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Start to space inflate position.cpp

It's a big file!

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix a performance bug in generate_move_if_legal

Use the pinned argument in pos.move_is_legal()

No functional change, simply use pos.move_is_legal() as
was meant to be.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Unify pinned and discovery checks code

Templates are our friends here. No functional change.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Refactor Position::pinned_pieces() to use templates

Also better document this interesting function.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

movegen: add SERIALIZE_MOVES and hides a bunch of loops

Only syntactic sugar, perhaps we should leave as is, anyhow...

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Another  generate_piece_moves() micro optimization

This time on the for loop.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix another template conversion bug in movegen

Hopefully the last one.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

movegen: prefer (*mlist++) to mlist[n++]

Teoretically faster, practically it helps to
removes some more lines.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix a bug in generate_pawn_captures()

Introduced in "movegen: Introduce generate_pawn_captures()"
when unifiying black and white functions.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fully templetize pawn move generators

A little bit more syntax heavuer but surely faster.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Micro optimize generate_piece_moves()

It is a time critical path. The biggest in move generation.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Another Intel warning sqeezed

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Quiet a warning on Intel compiler

Plus usual trailing whitespace noise.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Shrink arguments in move generation functions

Perhaps no speedup, but it is more readable.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix a bug in king discoveries checks

Introduced in "Add a generate_piece_checks() specialization for the king"

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Update copyright info

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Last little touches to movegen

No functional change.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Rename PawnOffsets in PawnParams

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Pawn move generator: dispatch at compile time

Instead of function pointers use templates to
dispatch shift operations.

It is more clear and possibly also faster because
branches are removed at compile time.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Start to templetize pawn move generators

Still very soft, we will see if compiler is
enough or we need more aggressive templetization.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Remove Positions::xxx_count() functions

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Remove white/black_pawn_attacks_square()

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Templetize Position::xxx_attacks_square()

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Remove Position::xxx_list() functions

No useful, only obfuscating.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Add a generate_piece_checks() specialization for the king

Also reshuffle the code a bit.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Remove white/black_pawn_attacks()

Unuseful syntactic sugar, obfuscates the
real code.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Prefer template to name decoration

This also allows faster code although bigger.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix a missing comma in BenchmarkPositions[]

An old bug introduced in 3e0dc9ee8477 almost
one month ago.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Position::is_ok()give more info on failed test

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix an assert due to a missing parentesis

Bitwise operators precedence issue here, was
causing an assert.

This is a fallout from recent patches.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Final semplification of generate_evasions()

Now it's readable!

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Introduce generate_piece_blocking_evasions()

Start to simplify generate_evasions

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Space inflate generate_evasions()

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Update copyright info

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Last little touches to movegen

No functional change.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Rename PawnOffsets in PawnParams

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Pawn move generator: dispatch at compile time

Instead of function pointers use templates to
dispatch shift operations.

It is more clear and possibly also faster because
branches are removed at compile time.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Start to templetize pawn move generators

Still very soft, we will see if compiler is
enough or we need more aggressive templetization.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Remove Positions::xxx_count() functions

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Remove white/black_pawn_attacks_square()

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Templetize Position::xxx_attacks_square()

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Remove Position::xxx_list() functions

No useful, only obfuscating.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Add a generate_piece_checks() specialization for the king

Also reshuffle the code a bit.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Remove white/black_pawn_attacks()

Unuseful syntactic sugar, obfuscates the
real code.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Prefer template to name decoration

This also allows faster code although bigger.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

Fix a missing comma in BenchmarkPositions[]

An old bug introduced in 3e0dc9ee8477 almost
one month ago.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>