[gnugo-devel] Owl persistent cache

[Daniel Bump]

Gunnar and I were watching the game:

jagm-gnugo-3.1.26-200202262313.sgf

available at http://www.lysator.liu.se/~gunnar/gnugo/nngs/
and Gunnar commented that the key to speeding up GNU Go
would be to speed up the owl code, for example by using
it less. At some time between move 50 and move 117, he
told me that the owl status of C6 was being recomputed
every move, though from moves 50-117 nothing had much
effect on that group. Probably the active area was
too big.

So an important speedup would be to improve the active
area of the owl persistent cache. The patch active_1_27.1
(not in the CVS yet) adds a debug flag -d0x200000 
(DEBUG_OWL_PERSISTENT_CACHE) that activates some disabled
debugging code that Gunnar had left in owl.c. I think
this gives a pretty good picture.

If you run this some you'll see that the active areas
do indeed look capable of improvement. I think
inventing a good new algorithm for this might be
an important thing right now.

Dan