Re: [gnugo-devel] Discussion about the new cache implementation

[Arend Bayer]

Inge wrote:

> Ok, that was it.  If nobody comes up with good objections, I think I
> will use the following plan:
>
> 1) Go back to the old type of hash value (two 32-bit ints).
> 2) Add more calls to the transposition table in other reading
>    functions than attack() and find_defense().
> 3) Establish some figures for the offset for owl nodes, semeai nodes
>    and connection nodes.
> 4) Implement the new cache scheme for these types of reading.
> 5) Remove the old cache.
>
> If anybody wants to help me, I would be especially grateful if you
> would do some work on point 3 above.  Or maybe suggest another way to
> solve the problem.

Sounds good. For no 3): If you want you can just always use the
_tactical_ reading nodes that were needed for a specific result. Most of
the time of owl reading is spent in the tactical reading code, and I guess
about half of the connection reading runtime, too. So that is a pretty good
figure to start with I'd say.

Arend