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>
They hide the underlying uniform function call with
no benefit.
A little bit more verbose but now is clear what happens.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
In generate_piece_moves() to be more uniform with other
functions. Unfortunatly the different number of calling arguments
do not allow us to easily integrate in generate_piece_moves()
template family.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Small code tidy up and a little optimization
to avoid calling generate_piece_blocking_evasions()
when blockSquares is empty (30% of cases).
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Now that we have removed sliders checkers attacks
from evasion's set we can simplyfy legality check.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Rewrite as in generate_piece_moves() using a for
loop instead of the slower serializing of the
bitboard with pop_1st_bit()
This will allow also to merge with generate_piece_moves()
when we will drop legality constrain on generate_evasions()
Generated moves are not changed, but are generated in a
different order, this changes the number of nodes at fixed
depth test.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Discovery check candidates are normally empty, so
avoid discovery checks generation in that common case.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Find squares attacked by slider checkers, we will
remove them from king evasions set so to avoid a couple
of cycles in the slow king evasions legality check loop.
Not a biggie, but now generate_evasions() is faster then
generate_non_captures(), before was slower.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
When we generate checks one case is missing: generation
of castling moves that give check to the opponent king.
This is a very rare case but anyway it is a case
and we can do this without slowing down the common
case of no castling checks.
So this is the patch.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
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>
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>
Introduced in "movegen: Introduce generate_pawn_captures()"
when unifiying black and white functions.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
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>
