Instead of classical flags, throw an
exception when we want to immediately halt
the search. Currently only one type
is used for both UCI stop and threads
cut off.
No functional change.
Because VALUE_NONE is 30002, it happens that
after a check the next move is never an improving
one.
After this patch bench signature is independent from
VALUE_NONE actual value.
bench: 4303194
Apply to LMR the same Eelco's idea
applied to move count pruning.
This is the result of a series of
attempts started by Thomas Kolarik.
Passed both short TC
LLR: 2.95 (-2.94, 2.94)
Total: 5675 W: 1241 L: 1117 D: 3317
And long TC:
LLR: 2.95 (-2.94, 2.94)
Total: 8748 W: 1689 L: 1539 D: 5520
bench: 4356801
Introduce ThreadBase struct that is search
agnostic and just handles low level stuff,
and derive all the other specialized classes
form here.
In particular TimerThread does not hinerits
anymore all the search related stuff from Thread.
Also some renaming while there.
Suggested by Steven Edwards
No functional change.
A big simplification and removing of useless code.
Finished at 50% both at short TC (with SPRT) than
at long TC at fixed number of games:
ELO: -0.14 +-3.4 (95%) LOS: 46.8%
Total: 15206 W: 2836 L: 2842 D: 9528
bench: 5059948
Here speed up is the name of the game.
Speed up is gained:
- Removing the useless enoughMaterial code
- Limiting trapped rook evaluation to where it counts
Tested at long TC:
LLR: 2.97 (-2.94,2.94)
Total: 10061 W: 1948 L: 1790 D: 6323
bench: 4558173
Thanks to Don, Miguel, Louis and the other people
of talkchess forum for the suggestion:
http://www.talkchess.com/forum/viewtopic.php?t=48612
Also sync polyglot.ini with current UCI options
No functional change.
Unfortunatly a reverse test at long TC failed:
master^ vs master
LLR: 1.37 (-2.94,2.94)
Total: 33682 W: 6294 L: 6071 D: 21317
So becuase short TC score is 50% there is a good
possibility patch is not scalable.
So revert it.
bench: 4507288
Do not use threat move to detect the condition. This
let us to retire the big allows() function.
Test at short TC was within 50% score:
LLR: -2.95 (-2.94,2.94)
Total: 38272 W: 7941 L: 7940 D: 22391
To be verified with reverse long TC
bench: 4191565
Here the main difference is that now we center
aspiration window on last returned score. This allows
to simplify handling of mate scores.
We have done a reversed SPRT tests, where we wanted to
verify if master is stronger than this patch.
Long TC: master vs this patch (reverse test)
LLR: -2.95 (-2.94,2.94)
Total: 37992 W: 7012 L: 6920 D: 24060
bench: 4507288
Temporary revert aspiration window patch
so to be visible to everybody: it will be
re-applied with next patch
No functional change (together with next one)
A simplification of the 'dangerous' definition.
Seems neutral at reverse test at long TC
master vs patch
LLR: -2.96 (-2.94,2.94)
Total: 16974 W: 3122 L: 3139 D: 10713
bench: 4689029
Function calloc() already initializes memory to
zero, so avoid calling clear() afterwards.
Also some renaming while there (inspired by DiscoCheck).
No functional change.
So when we are doing a LMR search at the parent ALL node.
This patch didn't prove stronger at 60" TC
LLR: -2.97 (-2.94,2.94)
Total: 22398 W: 4070 L: 4060 D: 14268
But, first, it scores at 50%, second (and most important for me) the opposite,
i.e. normal reduction when parent node is not reduced, seems very bad:
LLR: -2.95 (-2.94,2.94)
Total: 7036 W: 1446 L: 1534 D: 4056
According to Don, this idea of increased reduction of CUT nodes
works because if parent node is reduced, missing a cut-off due to
reduced depth search (meaning position is somehow tricky) forces
a full depth research at parent node, giving due insight in this
set of sensible positions.
IOW if we expect a node to fail-high at depth n, then we assume it
should fail-high also at depth n-1, if this doesn't happen it means
position is tricky enough to deserve a research at depth n+1.
bench: 4687419
We got a good result from this tweak, in line with
what was already found by Don Dailey.
At short TC:
LLR: 2.95 (-2.94,2.94)
Total: 13097 W: 2742 L: 2598 D: 7757
At long TC:
LLR: 2.97 (-2.94,2.94)
Total: 7281 W: 1408 L: 1265 D: 4608
bench: 5108393
Follow Don Dailey definition of cut/all node:
"If the previous node was a cut node, we consider this an ALL node.
The only exception is for PV nodes which are a special case of ALL nodes.
In the PVS framework, the first zero width window searched from a PV
node is by our definition a CUT node and if you have to do a re-search
then it is suddenly promoted to a PV nodes (as per PVS search) and only
then can the cut and all nodes swap positions. In other words, these
internal search failures can force the status of every node in the subtree
to swap if it propagates back to the last PV nodes."
http://talkchess.com/forum/viewtopic.php?topic_view=threads&p=519741&t=47577
With this definition we have an hit rate higher than 90% on:
if (!PvNode && depth > 4 * ONE_PLY)
dbg_hit_on_c(cutNode, (bestValue >= beta));
And an hit rate of just 28% on:
if (!PvNode && depth > 4 * ONE_PLY)
dbg_hit_on_c(!cutNode, (bestValue >= beta));
No functional change.
Fix was wrong becuase search starts from ss+1,
code is a bit tricky here, so rewrite in a way
to be more easy to read and understand.
Spotted by Eelco.
No functional change.
