Rever count_1s() optimizations

[stockfish] / src / bitboard.h

diff --git a/src/bitboard.h b/src/bitboard.h
index eceb845e643112d929c24a8a0b6236ab574fd589..1aefd8add5a623729652f70fc21a45076aacb5ee 100644 (file)
--- a/src/bitboard.h
+++ b/src/bitboard.h
@@ -8,6 +8,7 @@
   the Free Software Foundation, either version 3 of the License, or
   (at your option) any later version.
 
+
   Stockfish is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
@@ -112,8 +113,8 @@ extern const Bitboard RankBB[8];
 extern const Bitboard RelativeRankBB[2][8];
 extern const Bitboard InFrontBB[2][8];
 
-extern Bitboard SetMaskBB[64];
-extern Bitboard ClearMaskBB[64];
+extern Bitboard SetMaskBB[65];
+extern Bitboard ClearMaskBB[65];
 
 extern Bitboard StepAttackBB[16][64];
 extern Bitboard RayBB[64][8];
@@ -235,6 +236,19 @@ inline Bitboard in_front_bb(Color c, Square s) {
 }
 
 
+/// behind_bb() takes a color and a rank or square as input, and returns a
+/// bitboard representing all the squares on all ranks behind of the rank
+/// (or square), from the given color's point of view.
+
+inline Bitboard behind_bb(Color c, Rank r) {
+  return InFrontBB[opposite_color(c)][r];
+}
+
+inline Bitboard behind_bb(Color c, Square s) {
+  return in_front_bb(opposite_color(c), square_rank(s));
+}
+
+
 /// ray_bb() gives a bitboard representing all squares along the ray in a
 /// given direction from a given square.
 
@@ -381,24 +395,25 @@ inline int count_1s_max_15(Bitboard b) {
 
 inline int count_1s(Bitboard b) {
   unsigned w = unsigned(b >> 32), v = unsigned(b);
-  v = v - ((v >> 1) & 0x55555555);
-  w = w - ((w >> 1) & 0x55555555);
-  v = (v & 0x33333333) + ((v >> 2) & 0x33333333);
-  w = (w & 0x33333333) + ((w >> 2) & 0x33333333);
-  v = (v + (v >> 4)) & 0x0F0F0F0F;
-  w = (w + (w >> 4)) & 0x0F0F0F0F;
-  v = ((v+w) * 0x01010101) >> 24; // mul is fast on amd procs
-  return int(v);
+  v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
+  w -= (w >> 1) & 0x55555555;
+  v = ((v >> 2) & 0x33333333) + (v & 0x33333333); // 0-4 in 4 bits
+  w = ((w >> 2) & 0x33333333) + (w & 0x33333333);
+  v = ((v >> 4) + v) & 0x0F0F0F0F; // 0-8 in 8 bits
+  v += (((w >> 4) + w) & 0x0F0F0F0F);  // 0-16 in 8 bits
+  v *= 0x01010101; // mul is fast on amd procs
+  return int(v >> 24);
 }
 
 inline int count_1s_max_15(Bitboard b) {
   unsigned w = unsigned(b >> 32), v = unsigned(b);
-  v = v - ((v >> 1) & 0x55555555);
-  w = w - ((w >> 1) & 0x55555555);
-  v = (v & 0x33333333) + ((v >> 2) & 0x33333333);
-  w = (w & 0x33333333) + ((w >> 2) & 0x33333333);
-  v = ((v+w) * 0x11111111) >> 28;
-  return int(v);
+  v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
+  w -= (w >> 1) & 0x55555555;
+  v = ((v >> 2) & 0x33333333) + (v & 0x33333333); // 0-4 in 4 bits
+  w = ((w >> 2) & 0x33333333) + (w & 0x33333333);
+  v += w; // 0-8 in 4 bits
+  v *= 0x11111111;
+  return int(v >> 28);
 }
 
 #elif defined(BITCOUNT_SWAR_64)