X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=src%2Fbitcount.h;h=871247abcf4be72b16df1def8d04f97a6e6d8f52;hb=76024ac40ea45e44a8389cb4206d40b884e259a5;hp=4016063383938fe7fbcc9309ff9767a2bfec6a7c;hpb=e7d3a006cda7822517159cad0ef222eb7e46db00;p=stockfish

diff --git a/src/bitcount.h b/src/bitcount.h
index 40160633..871247ab 100644
--- a/src/bitcount.h
+++ b/src/bitcount.h
@@ -22,6 +22,12 @@
 #if !defined(BITCOUNT_H_INCLUDED)
 #define BITCOUNT_H_INCLUDED
 
+// To disable POPCNT support uncomment following line. You should do it only
+// in PGO compiling to exercise the default fallback path. Don't forget to
+// re-comment the line for the final optimized compile though ;-)
+//#define DISABLE_POPCNT_SUPPORT
+
+
 #include "bitboard.h"
 
 
@@ -48,7 +54,7 @@
 
 // Select type of intrinsic bit count instruction to use
 
-#if defined(_MSC_VER) // Microsoft compiler
+#if defined(_MSC_VER) && defined(_WIN64) // Microsoft compiler
 
 #include <intrin.h>
 
@@ -60,6 +66,7 @@ inline bool cpu_has_popcnt() {
 }
 
 #define POPCNT_INTRINSIC(x) __popcnt64(x)
+#define BITSCAN_INTRINSIC(idx, x) _BitScanForward64(idx, x)
 
 #elif defined(__INTEL_COMPILER) && (defined(__x86_64) || defined(_M_X64)) // Intel compiler
 
@@ -73,12 +80,14 @@ inline bool cpu_has_popcnt() {
 }
 
 #define POPCNT_INTRINSIC(x) _mm_popcnt_u64(x)
+#define BITSCAN_INTRINSIC(idx, x) _BitScanForward64(idx, x)
 
 #else // Safe fallback for unsupported compilers
 
 inline bool cpu_has_popcnt() { return false; }
 
-#define POPCNT_INTRINSIC(x) sw_count_1s(x)
+#define POPCNT_INTRINSIC(x) count_1s(x)
+#define BITSCAN_INTRINSIC(idx, x) count_1s(x) // dummy
 
 #endif
 
@@ -87,19 +96,19 @@ inline bool cpu_has_popcnt() { return false; }
 
 #if defined(BITCOUNT_LOOP)
 
-inline int sw_count_1s(Bitboard b) {
+inline int count_1s(Bitboard b) {
   int r;
   for(r = 0; b; r++, b &= b - 1);
   return r;
 }
 
-inline int sw_count_1s_max_15(Bitboard b) {
+inline int count_1s_max_15(Bitboard b) {
   return count_1s(b);
 }
 
 #elif defined(BITCOUNT_SWAR_32)
 
-inline int sw_count_1s(Bitboard b) {
+inline int count_1s(Bitboard b) {
   unsigned w = unsigned(b >> 32), v = unsigned(b);
   v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
   w -= (w >> 1) & 0x55555555;
@@ -111,7 +120,7 @@ inline int sw_count_1s(Bitboard b) {
   return int(v >> 24);
 }
 
-inline int sw_count_1s_max_15(Bitboard b) {
+inline int count_1s_max_15(Bitboard b) {
   unsigned w = unsigned(b >> 32), v = unsigned(b);
   v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
   w -= (w >> 1) & 0x55555555;
@@ -124,7 +133,7 @@ inline int sw_count_1s_max_15(Bitboard b) {
 
 #elif defined(BITCOUNT_SWAR_64)
 
-inline int sw_count_1s(Bitboard b) {
+inline int count_1s(Bitboard b) {
   b -= ((b>>1) & 0x5555555555555555ULL);
   b = ((b>>2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
   b = ((b>>4) + b) & 0x0F0F0F0F0F0F0F0FULL;
@@ -132,7 +141,7 @@ inline int sw_count_1s(Bitboard b) {
   return int(b >> 56);
 }
 
-inline int sw_count_1s_max_15(Bitboard b) {
+inline int count_1s_max_15(Bitboard b) {
   b -= (b>>1) & 0x5555555555555555ULL;
   b = ((b>>2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
   b *= 0x1111111111111111ULL;
@@ -149,19 +158,53 @@ inline int sw_count_1s_max_15(Bitboard b) {
 template<bool UseIntrinsic>
 inline int count_1s(Bitboard b) {
 
-  return UseIntrinsic ? POPCNT_INTRINSIC(b) : sw_count_1s(b);
+  return UseIntrinsic ? POPCNT_INTRINSIC(b) : count_1s(b);
 }
 
 template<bool UseIntrinsic>
 inline int count_1s_max_15(Bitboard b) {
 
-  return UseIntrinsic ? POPCNT_INTRINSIC(b) : sw_count_1s_max_15(b);
+  return UseIntrinsic ? POPCNT_INTRINSIC(b) : count_1s_max_15(b);
 }
 
 
-// Global variable initialized at startup that is set to true if
-// CPU on which application runs support POPCNT intrinsic.
+/// pop_1st_bit() finds and clears the least significant nonzero bit in a
+/// nonzero bitboard. If template parameter is true an intrinsic is called,
+/// otherwise we fallback on a software implementation.
+
+template<bool UseIntrinsic>
+inline Square pop_1st_bit(Bitboard *b) {
+
+  return pop_1st_bit(b);
+}
+
+template<>
+inline Square pop_1st_bit<true>(Bitboard *b) {
+
+  unsigned long idx;
+  Bitboard bb = *b;
+  BITSCAN_INTRINSIC(&idx, bb);
+  *b &= (bb - 1);
+  return Square(idx);
+}
+
 
+// Global variable initialized at startup that is set to true if
+// CPU on which application runs supports POPCNT intrinsic. Unless
+// DISABLE_POPCNT_SUPPORT is defined.
+#if defined(DISABLE_POPCNT_SUPPORT)
+const bool CpuHasPOPCNT = false;
+#else
 const bool CpuHasPOPCNT = cpu_has_popcnt();
+#endif
+
+
+// Global variable used to print info about the use of 64 optimized
+// functions to verify that a 64bit compile has been correctly built.
+#if defined(BITCOUNT_SWAR_64)
+const bool CpuHas64BitPath = true;
+#else
+const bool CpuHas64BitPath = false;
+#endif
 
 #endif // !defined(BITCOUNT_H_INCLUDED)