Yet another pop_1st_bit() optimization

[stockfish] / src / bitboard.cpp

diff --git a/src/bitboard.cpp b/src/bitboard.cpp
index 458f9318f9d00d0d6c3b716c6268a78d7caef9de..89065c49fe81eb11d0328e08b121cb0018c23f77 100644 (file)
--- a/src/bitboard.cpp
+++ b/src/bitboard.cpp
@@ -21,6 +21,16 @@
 //// Includes
 ////
 
+#ifdef _MSC_VER
+    #include <intrin.h>
+    #ifdef _WIN64
+        #pragma intrinsic(_BitScanForward64)
+    #else
+        #pragma intrinsic(_BitScanForward)
+    #endif
+    #define USING_INTRINSICS
+#endif
+
 #include <iostream>
 
 #include "bitboard.h"
@@ -274,7 +284,7 @@ namespace {
 #if defined(USE_COMPACT_ROOK_ATTACKS)
   void init_file_and_rank_attacks();
 #endif
-};
+}
 
 
 ////
@@ -339,20 +349,39 @@ Square first_1(Bitboard b) {
 /// pop_1st_bit() finds and clears the least significant nonzero bit in a
 /// nonzero bitboard.
 
-#if defined(USE_32BIT_ATTACKS) && defined(_WIN32)
+#if defined(USE_32BIT_ATTACKS)
+
+// Use type-punning
+union b_union {
+
+    Bitboard b;
+    struct {
+        uint32_t l;
+        uint32_t h;
+    };
+};
 
+// WARNING: Needs -fno-strict-aliasing compiler option
 Square pop_1st_bit(Bitboard *bb) {
 
-  uint32_t  a = uint32_t(*bb);
-  uint32_t* ptr = a ? (uint32_t*)bb : (uint32_t*)bb + 1; // Little endian only?
-  uint32_t  b = a ? a : *ptr;
-  uint32_t  c = ~(b ^ (b - 1));
+  b_union u;
+  uint32_t b;
 
-  *ptr = b & c; // clear the bit
-  if (a)
-     c = ~c;
+  u.b = *bb;
 
-  return Square(BitTable[(c * 0x783a9b23) >> 26]);
+  if (u.l)
+  {
+      b = u.l;
+      *((uint32_t*)bb) = b & (b - 1);
+      b ^= (b - 1);
+  }
+  else
+  {
+      b = u.h;
+      *((uint32_t*)bb+1) = b & (b - 1); // Little endian only?
+      b = ~(b ^ (b - 1));
+  }
+  return Square(BitTable[(b * 0x783a9b23) >> 26]);
 }
 
 #else