🚀 go-pugleaf

At present, the KILLT and HISTORY heuristics overlapped (see genmoves.c).
I would like to suggest that these two heuristics be separated.  Let
HISTORY only refer to history data and nothing to do with killer moves.
What this means is that in MoveList() and VMoveList(), those #ifdef
HISTORY blocks should be commented out (or removed).

This makes it necessary to include -DKILLT in the Makefile to activate
killer moves heuristics.  The history heuristic can then be activated 
using -DHISTORY and this will not interfere with the killer heuristics.

I also think that the current implementation of HISTORY is buggy.  After
some experimentation, this is what I would like to suggest.  In search(),
at the bottom where the history[] array is being calculated, replace

if (history[j] < HISTORYLIM)
  history[j] += (unsigned short) 1 << depth;

with

if (history[j] < ((unsigned short) 1 << depth))
   history[j] = ((unsigned short) 1 << depth);

So instead of incrementing history[j] everytime, we set history[j] to the
maximum value that is seen.  In a few test positions, this gives much
better cuts than the original code.


Kong Sian