Also search null move in QS iteration of deeper nodes

The QS iteration of deeper nodes is intended to replace sorting
the captures in front of the 1-ply iteration. Not to generate an actual
QS result. It is thus not a good idea to stand pat or take a stand-pat
cutoff in this iteration, as this could mean we start the ply-1 iteration
blindly (and thus likely with pointless non-captures) if null-move turns
out to be worse than stand-pat (failing low) because of a threat.
 So we now do null move not based on iteration number, but based on
total node depth. Except that in the pre-QS iteration it is still suppressed.

diff --git a/fairymax.c b/fairymax.c
index 50a6f79..f9cf637 100644 (file)
--- a/fairymax.c
+++ b/fairymax.c
@@ -163,7 +163,7 @@ int k,q,l,e,ev,E,z,n;    /* (q,l)=window, e=current eval. score, E=e.p. sqr.*/
    (K=X,L=Y&~S,Score=m,d=3)))                  /* time's up: go do best    */\r
  {x=B=X;                                       /* start scan at prev. best */\r
   h=Y&S;                                       /* request try noncastl. 1st*/\r
-  P=d>2&&l+I?D(16-k,-l,1-l,-e,-ev,2*S,2*S,d-3):I;  /* search null move     */\r
+  P=n>2&d>1&&l+I?D(16-k,-l,1-l,-e,-ev,2*S,2*S,d-3):I;  /* search null move */\r
   m=-P<l|R<5?d-2?-I:ev:-P;  /*** prune if > beta  unconsidered:static eval */\r
   SHAMAX( if(pl[k]<=1&pl[16-k]>1)m=I-1; )      /* bare king loses          */\r
   N++;                                         /* node count (for timing)  */\r