Killer moves
Moderator: Ras
-
Fabio Gobbato
- Posts: 219
- Joined: Fri Apr 11, 2014 10:45 am
- Full name: Fabio Gobbato
Killer moves
Which is an average ratio between a killer move probe and a cut off during the search?
-
spambanane
- Posts: 22
- Joined: Sun Jun 17, 2012 9:45 am
Re: Killer moves
here some numbers, bratko-kopec test, 2s each:
killer 1 available: 3,244,389, killer 1 pseudo legal: 1,641,169
killer 2 available: 3,209,404, killer 2 pseudo legal: 1,401,643
cut offs: 1,302,966
i don't try killers from depth-2. before killers i try hash move and captures (mvv/lva sorted).
killers gave my engine a significant speed up. only thing i have to investigate is a huge slow down in "lasker-reichhelm" position.
heiko
killer 1 available: 3,244,389, killer 1 pseudo legal: 1,641,169
killer 2 available: 3,209,404, killer 2 pseudo legal: 1,401,643
cut offs: 1,302,966
i don't try killers from depth-2. before killers i try hash move and captures (mvv/lva sorted).
killers gave my engine a significant speed up. only thing i have to investigate is a huge slow down in "lasker-reichhelm" position.
heiko
-
tpetzke
- Posts: 686
- Joined: Thu Mar 03, 2011 4:57 pm
- Location: Germany
Re: Killer moves
This indicates a potential problem in the replacement scheme for the transposition table.killers gave my engine a significant speed up. only thing i have to investigate is a huge slow down in "lasker-reichhelm" position.
Thomas...
-
spambanane
- Posts: 22
- Joined: Sun Jun 17, 2012 9:45 am
Re: Killer moves
hi thomas,
thanks for the hint! does this knowlegde come from own experience? i'd never thought it would be a tt problem. as i can't see any relation between killers and tt. only bug in my tt i know of is that i store mate scores in a wrong way.
tt replacement scheme is "depth first + 4 buckets", btw.
thanks for the hint! does this knowlegde come from own experience? i'd never thought it would be a tt problem. as i can't see any relation between killers and tt. only bug in my tt i know of is that i store mate scores in a wrong way.
tt replacement scheme is "depth first + 4 buckets", btw.
-
tpetzke
- Posts: 686
- Joined: Thu Mar 03, 2011 4:57 pm
- Location: Germany
Re: Killer moves
Hi,
the FINE70 position is usually taken as a stress test for the transposition table as it reveals potential problems.
With a good TT this position should be resolved in almost no time.
A potential problem in the TT implementation here is that you keep entries with higher draft for to long in the table.
If you want to store a position in the TT and find you already have it in with higher draft, you should replace it anyway with your lower draft entry. The higher draft entry was useless otherwise you would not have come to the point where you want to store a lower draft entry.
The killers don't have directly something to do with it, except that they change the move order and the changed move order can then interfere with the TT.
Thomas...
the FINE70 position is usually taken as a stress test for the transposition table as it reveals potential problems.
With a good TT this position should be resolved in almost no time.
A potential problem in the TT implementation here is that you keep entries with higher draft for to long in the table.
If you want to store a position in the TT and find you already have it in with higher draft, you should replace it anyway with your lower draft entry. The higher draft entry was useless otherwise you would not have come to the point where you want to store a lower draft entry.
The killers don't have directly something to do with it, except that they change the move order and the changed move order can then interfere with the TT.
Thomas...