Re: [gnugo-devel] more board.c optimizations

[Paul Pogonyshev]

Arend wrote:
> Out of curiousity: Did you measure how many of the calls to
> approxlib() and accuratelib() can be answered from cache? (You could
> generate a profile and compare the number of calls to approxlib()
> and do_approxlib().)

i used a crude method which suggested 20-30% cache effectiveness. not
too high, but enough to give a decent speedup.

> If the number is still low, say well beyond 50%, you could try
> increasing the size of the cache (e.g. always store the latest 3 entries
> or s.th.  similar).

i have tried this (we also had a discussion with Arend about the matter
on nngs). version with large cache is faster than cvs, but slower than
version with simple cache (i.e. original patch). i'm not sure why, but
probably there are not many duplicated calls to approxlib() and
accuratelib() and hence there's no need in complicated caching scheme.

Paul