2 Stockfish, a UCI chess playing engine derived from Glaurung 2.1
3 Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
4 Copyright (C) 2008-2009 Marco Costalba
6 Stockfish is free software: you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation, either version 3 of the License, or
9 (at your option) any later version.
12 Stockfish is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>.
22 #if !defined(BITCOUNT_H_INCLUDED)
23 #define BITCOUNT_H_INCLUDED
25 // To disable POPCNT support uncomment following line. You should do it only
26 // in PGO compiling to exercise the default fallback path. Don't forget to
27 // re-comment the line for the final optimized compile though ;-)
28 //#define DISABLE_POPCNT_SUPPORT
34 // Select type of software bit count function to use
36 #if !defined(AUTO_CONFIGURATION) || defined(IS_64BIT)
38 //#define USE_COMPACT_ROOK_ATTACKS
39 //#define USE_32BIT_ATTACKS
40 #define USE_FOLDED_BITSCAN
42 #define BITCOUNT_SWAR_64
43 //#define BITCOUNT_SWAR_32
44 //#define BITCOUNT_LOOP
48 #define USE_32BIT_ATTACKS
49 #define USE_FOLDED_BITSCAN
50 #define BITCOUNT_SWAR_32
55 // Select type of intrinsic bit count instruction to use
57 #if defined(_MSC_VER) && defined(_WIN64) // Microsoft compiler
61 inline bool cpu_has_popcnt() {
63 int CPUInfo[4] = {-1};
64 __cpuid(CPUInfo, 0x00000001);
65 return (CPUInfo[2] >> 23) & 1;
68 #define POPCNT_INTRINSIC(x) __popcnt64(x)
69 #define BITSCAN_INTRINSIC(idx, x) _BitScanForward64(idx, x)
71 #elif defined(__INTEL_COMPILER) && (defined(__x86_64) || defined(_M_X64)) // Intel compiler
73 #include <nmmintrin.h>
75 inline bool cpu_has_popcnt() {
77 int CPUInfo[4] = {-1};
78 __cpuid(CPUInfo, 0x00000001);
79 return (CPUInfo[2] >> 23) & 1;
82 #define POPCNT_INTRINSIC(x) _mm_popcnt_u64(x)
83 #define BITSCAN_INTRINSIC(idx, x) _BitScanForward64(idx, x)
85 #else // Safe fallback for unsupported compilers
87 inline bool cpu_has_popcnt() { return false; }
89 #define POPCNT_INTRINSIC(x) count_1s(x)
90 #define BITSCAN_INTRINSIC(idx, x) count_1s(x) // dummy
95 /// Software implementation of bit count functions
97 #if defined(BITCOUNT_LOOP)
99 inline int count_1s(Bitboard b) {
101 for(r = 0; b; r++, b &= b - 1);
105 inline int count_1s_max_15(Bitboard b) {
109 #elif defined(BITCOUNT_SWAR_32)
111 inline int count_1s(Bitboard b) {
112 unsigned w = unsigned(b >> 32), v = unsigned(b);
113 v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
114 w -= (w >> 1) & 0x55555555;
115 v = ((v >> 2) & 0x33333333) + (v & 0x33333333); // 0-4 in 4 bits
116 w = ((w >> 2) & 0x33333333) + (w & 0x33333333);
117 v = ((v >> 4) + v) & 0x0F0F0F0F; // 0-8 in 8 bits
118 v += (((w >> 4) + w) & 0x0F0F0F0F); // 0-16 in 8 bits
119 v *= 0x01010101; // mul is fast on amd procs
123 inline int count_1s_max_15(Bitboard b) {
124 unsigned w = unsigned(b >> 32), v = unsigned(b);
125 v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
126 w -= (w >> 1) & 0x55555555;
127 v = ((v >> 2) & 0x33333333) + (v & 0x33333333); // 0-4 in 4 bits
128 w = ((w >> 2) & 0x33333333) + (w & 0x33333333);
129 v += w; // 0-8 in 4 bits
134 #elif defined(BITCOUNT_SWAR_64)
136 inline int count_1s(Bitboard b) {
137 b -= ((b>>1) & 0x5555555555555555ULL);
138 b = ((b>>2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
139 b = ((b>>4) + b) & 0x0F0F0F0F0F0F0F0FULL;
140 b *= 0x0101010101010101ULL;
144 inline int count_1s_max_15(Bitboard b) {
145 b -= (b>>1) & 0x5555555555555555ULL;
146 b = ((b>>2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
147 b *= 0x1111111111111111ULL;
154 /// count_1s() counts the number of nonzero bits in a bitboard.
155 /// If template parameter is true an intrinsic is called, otherwise
156 /// we fallback on a software implementation.
158 template<bool UseIntrinsic>
159 inline int count_1s(Bitboard b) {
161 return UseIntrinsic ? POPCNT_INTRINSIC(b) : count_1s(b);
164 template<bool UseIntrinsic>
165 inline int count_1s_max_15(Bitboard b) {
167 return UseIntrinsic ? POPCNT_INTRINSIC(b) : count_1s_max_15(b);
171 /// pop_1st_bit() finds and clears the least significant nonzero bit in a
172 /// nonzero bitboard. If template parameter is true an intrinsic is called,
173 /// otherwise we fallback on a software implementation.
175 template<bool UseIntrinsic>
176 inline Square pop_1st_bit(Bitboard *b) {
178 return pop_1st_bit(b);
182 inline Square pop_1st_bit<true>(Bitboard *b) {
186 BITSCAN_INTRINSIC(&idx, bb);
192 // Global variable initialized at startup that is set to true if
193 // CPU on which application runs supports POPCNT intrinsic. Unless
194 // DISABLE_POPCNT_SUPPORT is defined.
195 #if defined(DISABLE_POPCNT_SUPPORT)
196 const bool CpuHasPOPCNT = false;
198 const bool CpuHasPOPCNT = cpu_has_popcnt();
202 // Global variable used to print info about the use of 64 optimized
203 // functions to verify that a 64bit compile has been correctly built.
204 #if defined(BITCOUNT_SWAR_64)
205 const bool CpuHas64BitPath = true;
207 const bool CpuHas64BitPath = false;
210 #endif // !defined(BITCOUNT_H_INCLUDED)