X-Git-Url: https://git.sesse.net/?p=stockfish;a=blobdiff_plain;f=src%2Fbitcount.h;h=a69ad8e8418535fbe01645368800815d19491b9f;hp=0992c00625e084ecc1fa78110ccef0c04a500166;hb=9204a60dbbebf7e319bb588acc91bf691a3ede9a;hpb=d4876dc96395f5592bfbc25b2eca2360db0655e6
diff --git a/src/bitcount.h b/src/bitcount.h
index 0992c006..a69ad8e8 100644
--- a/src/bitcount.h
+++ b/src/bitcount.h
@@ -1,7 +1,7 @@
/*
Stockfish, a UCI chess playing engine derived from Glaurung 2.1
Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
- Copyright (C) 2008-2010 Marco Costalba, Joona Kiiski, Tord Romstad
+ Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
Stockfish is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
@@ -18,108 +18,89 @@
along with this program. If not, see .
*/
-
#if !defined(BITCOUNT_H_INCLUDED)
#define BITCOUNT_H_INCLUDED
+#include
#include "types.h"
enum BitCountType {
- CNT64,
- CNT64_MAX15,
- CNT32,
- CNT32_MAX15,
- CNT_POPCNT
+ CNT_64,
+ CNT_64_MAX15,
+ CNT_32,
+ CNT_32_MAX15,
+ CNT_HW_POPCNT
};
-/// count_1s() counts the number of nonzero bits in a bitboard.
-/// We have different optimized versions according if platform
-/// is 32 or 64 bits, and to the maximum number of nonzero bits.
-/// We also support hardware popcnt instruction. See Readme.txt
-/// on how to pgo compile with popcnt support.
-template inline int count_1s(Bitboard);
+/// Determine at compile time the best popcount<> specialization according if
+/// platform is 32 or 64 bits, to the maximum number of nonzero bits to count
+/// and if hardware popcnt instruction is available.
+const BitCountType Full = HasPopCnt ? CNT_HW_POPCNT : Is64Bit ? CNT_64 : CNT_32;
+const BitCountType Max15 = HasPopCnt ? CNT_HW_POPCNT : Is64Bit ? CNT_64_MAX15 : CNT_32_MAX15;
+
+
+/// popcount() counts the number of nonzero bits in a bitboard
+template inline int popcount(Bitboard);
template<>
-inline int count_1s(Bitboard b) {
- b -= ((b>>1) & 0x5555555555555555ULL);
- b = ((b>>2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
- b = ((b>>4) + b) & 0x0F0F0F0F0F0F0F0FULL;
- b *= 0x0101010101010101ULL;
- return int(b >> 56);
+inline int popcount(Bitboard b) {
+ b -= (b >> 1) & 0x5555555555555555ULL;
+ b = ((b >> 2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
+ b = ((b >> 4) + b) & 0x0F0F0F0F0F0F0F0FULL;
+ return (b * 0x0101010101010101ULL) >> 56;
}
template<>
-inline int count_1s(Bitboard b) {
- b -= (b>>1) & 0x5555555555555555ULL;
- b = ((b>>2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
- b *= 0x1111111111111111ULL;
- return int(b >> 60);
+inline int popcount(Bitboard b) {
+ b -= (b >> 1) & 0x5555555555555555ULL;
+ b = ((b >> 2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
+ return (b * 0x1111111111111111ULL) >> 60;
}
template<>
-inline int count_1s(Bitboard b) {
+inline int popcount(Bitboard b) {
unsigned w = unsigned(b >> 32), v = unsigned(b);
- 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);
+ 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 + (w >> 4) + w) & 0x0F0F0F0F;
+ return (v * 0x01010101) >> 24;
}
template<>
-inline int count_1s(Bitboard b) {
+inline int popcount(Bitboard b) {
unsigned w = unsigned(b >> 32), v = unsigned(b);
- 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);
+ 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);
+ return ((v + w) * 0x11111111) >> 28;
}
template<>
-inline int count_1s(Bitboard b) {
+inline int popcount(Bitboard b) {
+
#if !defined(USE_POPCNT)
- return int(b != 0); // Avoid 'b not used' warning
-#elif defined(_MSC_VER)
- return __popcnt64(b);
-#elif defined(__GNUC__)
- unsigned long ret;
- __asm__("popcnt %1, %0" : "=r" (ret) : "r" (b));
- return ret;
-#endif
-}
+ assert(false);
+ return b != 0; // Avoid 'b not used' warning
-/// cpu_has_popcnt() detects support for popcnt instruction at runtime
-inline bool cpu_has_popcnt() {
+#elif defined(_MSC_VER) && defined(__INTEL_COMPILER)
- int CPUInfo[4] = {-1};
- __cpuid(CPUInfo, 0x00000001);
- return (CPUInfo[2] >> 23) & 1;
-}
+ return _mm_popcnt_u64(b);
+
+#elif defined(_MSC_VER)
+ return (int)__popcnt64(b);
-/// CpuHasPOPCNT is a global constant initialized at startup that
-/// is set to true if CPU on which application runs supports popcnt
-/// hardware instruction. Unless USE_POPCNT is not defined.
-#if defined(USE_POPCNT)
-const bool CpuHasPOPCNT = cpu_has_popcnt();
#else
-const bool CpuHasPOPCNT = false;
-#endif
+ unsigned long ret;
+ __asm__("popcnt %1, %0" : "=r" (ret) : "r" (b));
+ return ret;
-/// CpuIs64Bit is a global constant initialized at compile time that
-/// is set to true if CPU on which application runs is a 64 bits.
-#if defined(IS_64BIT)
-const bool CpuIs64Bit = true;
-#else
-const bool CpuIs64Bit = false;
#endif
+}
#endif // !defined(BITCOUNT_H_INCLUDED)