Better MultiPV performance

Allows better use of the TT for multiPV, resulting in greater depth and
thus strength.

src/search.cpp

@@ -589,7 +589,11 @@ Value Search::Worker::search(
             return alpha;
     }
     else
+    {
         thisThread->rootDelta = beta - alpha;
+        if (thisThread->pvIdx > 0 && depth > 3)
+            depth -= 2;
+    }
 
     assert(0 <= ss->ply && ss->ply < MAX_PLY);
 
@@ -614,8 +618,13 @@ Value Search::Worker::search(
     if (!excludedMove)
         ss->ttPv = PvNode || (ss->ttHit && tte->is_pv());
 
+    bool multiTT =    thisThread->pvIdx > 0
+                   && !rootNode
+                   && !(ss->ply & 1)
+                   && (tte->bound() & BOUND_LOWER);
+
     // At non-PV nodes we check for an early TT cutoff
-    if (!PvNode && !excludedMove && tte->depth() > depth
+    if ((!PvNode || multiTT) && !excludedMove && tte->depth() > depth
         && ttValue != VALUE_NONE  // Possible in case of TT access race or if !ttHit
         && (tte->bound() & (ttValue >= beta ? BOUND_LOWER : BOUND_UPPER)))
     {
@@ -1445,8 +1454,12 @@ Value Search::Worker::qsearch(Position& pos, Stack* ss, Value alpha, Value beta,
     ttMove  = ss->ttHit ? tte->move() : Move::none();
     pvHit   = ss->ttHit && tte->is_pv();
 
+    bool multiTT =    thisThread->pvIdx > 0
+                   && !(ss->ply & 1)
+                   && (tte->bound() & BOUND_LOWER);
+
     // At non-PV nodes we check for an early TT cutoff
-    if (!PvNode && tte->depth() >= ttDepth
+    if ((!PvNode || multiTT) && tte->depth() >= ttDepth
         && ttValue != VALUE_NONE  // Only in case of TT access race or if !ttHit
         && (tte->bound() & (ttValue >= beta ? BOUND_LOWER : BOUND_UPPER)))
         return ttValue;