1/*
2 Stockfish, a UCI chess playing engine derived from Glaurung 2.1
3 Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
4 Copyright (C) 2008-2015 Marco Costalba, Joona Kiiski, Tord Romstad
5 Copyright (C) 2015-2019 Marco Costalba, Joona Kiiski, Gary Linscott, Tord Romstad
6
7 Stockfish is free software: you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation, either version 3 of the License, or
10 (at your option) any later version.
11
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.
16
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/>.
19*/
20
21#include <cstring> // For std::memset
22#include <iostream>
23#include <thread>
24
25#include "bitboard.h"
26#include "misc.h"
27#include "thread.h"
28#include "tt.h"
29#include "uci.h"
30
31TranspositionTable TT; // Our global transposition table
32
33/// TTEntry::save populates the TTEntry with a new node's data, possibly
34/// overwriting an old position. Update is not atomic and can be racy.
35
36void TTEntry::save(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev) {
37
38 assert(d / ONE_PLY * ONE_PLY == d);
39
40 // Preserve any existing move for the same position
41 if (m || (k >> 48) != key16)
42 move16 = (uint16_t)m;
43
44 // Overwrite less valuable entries
45 if ( (k >> 48) != key16
46 ||(d - DEPTH_OFFSET) / ONE_PLY > depth8 - 4
47 || b == BOUND_EXACT)
48 {
49 assert((d - DEPTH_OFFSET) / ONE_PLY >= 0);
50
51 key16 = (uint16_t)(k >> 48);
52 value16 = (int16_t)v;
53 eval16 = (int16_t)ev;
54 genBound8 = (uint8_t)(TT.generation8 | uint8_t(pv) << 2 | b);
55 depth8 = (uint8_t)((d - DEPTH_OFFSET) / ONE_PLY);
56 }
57}
58
59
60/// TranspositionTable::resize() sets the size of the transposition table,
61/// measured in megabytes. Transposition table consists of a power of 2 number
62/// of clusters and each cluster consists of ClusterSize number of TTEntry.
63
64void TranspositionTable::resize(size_t mbSize) {
65
66 Threads.main()->wait_for_search_finished();
67
68 clusterCount = mbSize * 1024 * 1024 / sizeof(Cluster);
69
70 free(mem);
71 mem = malloc(clusterCount * sizeof(Cluster) + CacheLineSize - 1);
72
73 if (!mem)
74 {
75 std::cerr << "Failed to allocate " << mbSize
76 << "MB for transposition table." << std::endl;
77 exit(EXIT_FAILURE);
78 }
79
80 table = (Cluster*)((uintptr_t(mem) + CacheLineSize - 1) & ~(CacheLineSize - 1));
81 clear();
82}
83
84
85/// TranspositionTable::clear() initializes the entire transposition table to zero,
86// in a multi-threaded way.
87
88void TranspositionTable::clear() {
89
90 std::vector<std::thread> threads;
91
92 for (size_t idx = 0; idx < Options["Threads"]; ++idx)
93 {
94 threads.emplace_back([this, idx]() {
95
96 // Thread binding gives faster search on systems with a first-touch policy
97 if (Options["Threads"] > 8)
98 WinProcGroup::bindThisThread(idx);
99
100 // Each thread will zero its part of the hash table
101 const size_t stride = clusterCount / Options["Threads"],
102 start = stride * idx,
103 len = idx != Options["Threads"] - 1 ?
104 stride : clusterCount - start;
105
106 std::memset(&table[start], 0, len * sizeof(Cluster));
107 });
108 }
109
110 for (std::thread& th: threads)
111 th.join();
112}
113
114/// TranspositionTable::probe() looks up the current position in the transposition
115/// table. It returns true and a pointer to the TTEntry if the position is found.
116/// Otherwise, it returns false and a pointer to an empty or least valuable TTEntry
117/// to be replaced later. The replace value of an entry is calculated as its depth
118/// minus 8 times its relative age. TTEntry t1 is considered more valuable than
119/// TTEntry t2 if its replace value is greater than that of t2.
120
121TTEntry* TranspositionTable::probe(const Key key, bool& found) const {
122
123 TTEntry* const tte = first_entry(key);
124 const uint16_t key16 = key >> 48; // Use the high 16 bits as key inside the cluster
125
126 for (int i = 0; i < ClusterSize; ++i)
127 if (!tte[i].key16 || tte[i].key16 == key16)
128 {
129 tte[i].genBound8 = uint8_t(generation8 | (tte[i].genBound8 & 0x7)); // Refresh
130
131 return found = (bool)tte[i].key16, &tte[i];
132 }
133
134 // Find an entry to be replaced according to the replacement strategy
135 TTEntry* replace = tte;
136 for (int i = 1; i < ClusterSize; ++i)
137 // Due to our packed storage format for generation and its cyclic
138 // nature we add 263 (256 is the modulus plus 7 to keep the unrelated
139 // lowest three bits from affecting the result) to calculate the entry
140 // age correctly even after generation8 overflows into the next cycle.
141 if ( replace->depth8 - ((263 + generation8 - replace->genBound8) & 0xF8)
142 > tte[i].depth8 - ((263 + generation8 - tte[i].genBound8) & 0xF8))
143 replace = &tte[i];
144
145 return found = false, replace;
146}
147
148
149/// TranspositionTable::hashfull() returns an approximation of the hashtable
150/// occupation during a search. The hash is x permill full, as per UCI protocol.
151
152int TranspositionTable::hashfull() const {
153
154 int cnt = 0;
155 for (int i = 0; i < 1000 / ClusterSize; ++i)
156 for (int j = 0; j < ClusterSize; ++j)
157 cnt += (table[i].entry[j].genBound8 & 0xF8) == generation8;
158
159 return cnt * 1000 / (ClusterSize * (1000 / ClusterSize));
160}
161