Re: [gnugo-devel] Default reading cache size / Questions

[Gunnar Farnebäck]

Alain wrote:
> i tried various sizes for the -M parameter and it seems to have
> nearly no effect on the speed of the engine while the param is in
> the range 1 -> 512 (little slow down for huge cache)
> 
> i tried very small size 0.1 , 0.01 and 0.001 there is a significant
> slowdown for 0.001 (twice slower) but nothing terrible for 0.1 or
> 0.01

I don't think there has been any systematic investigations of the
effects of cache size after the reading cache was completely
reimplemented around 3.5.5. It is possible that a significantly
smaller cache than before would be adequate as default. Presumably a
large cache is more useful in games which require much reading. You
could for example try to replay regression/games/FSF-neurogo.sgf and
increase the level to 13 or so, and see whether that behaves
differently.

> Here is my question :
> in genmove, at each move during initialisation, there is a call to
> reset_engine() and it seems that it clears everything , so the cache
> is only used during one "move generation", and does not survive once
> the move is played.

The information from the previous move either doesn't match the
current full board or doesn't read to sufficient depth any longer so
everything needs to be recomputed anyway.

> i naively thought that the cache would persist during all the game. 
> Am i correct, or did i miss some important thing ? 

That's what the persistent cache in persistent.c is for. This cache
uses a heuristic active area to determine how long a result can be
trusted. 

/Gunnar