Re: [gnugo-devel] Improvement of the hash table in GNU go

[Inge Wallin]

Gunnar wrote:
> Inge wrote:
> > What I propose is a very simple but powerful change: We introduce a
> > new table of random number for each reading function that we want to
> > cache the result of.  Then we use these tables to create a unique hash
> > value for a combination of position and read function on a certain
> > target.  That way we get a single type of entries and can get rid of a
> > lot of the current complications.
> > 
> > Take for example attack().  If we call attack(A1), and want to find
> > out if the result of this reading is in the table we do the following:
> 
> A simpler alternative would be to have only one board-sized array with
> random numbers for each vertex and one random number for each routine,
> then xor the routine and the vertex numbers. This also has the
> advantage of more easily scaling to two vertices, which both semeai
> and connection reading need.
> 
> The same can of course be done with komaster, kom_pos, and remaining
> depth, which would comprise all current input data.

I was probably a bit unclear in my description, because this is what I
meant from the beginning.

      -Inge