/*
Stockfish, a UCI chess playing engine derived from Glaurung 2.1
- Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
- Copyright (C) 2008-2014 Marco Costalba, Joona Kiiski, Tord Romstad
+ Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
Stockfish is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
+#include "tt.h"
+
+#include <cassert>
+#include <cstdlib>
#include <cstring>
#include <iostream>
+#include <thread>
+#include <vector>
-#include "bitboard.h"
-#include "tt.h"
-
-TranspositionTable TT; // Our global transposition table
+#include "misc.h"
+#include "thread.h"
+#include "uci.h"
+namespace Stockfish {
-/// TranspositionTable::resize() sets the size of the transposition table,
-/// measured in megabytes. Transposition table consists of a power of 2 number
-/// of clusters and each cluster consists of ClusterSize number of TTEntry.
+TranspositionTable TT; // Our global transposition table
-void TranspositionTable::resize(uint64_t mbSize) {
+// Populates the TTEntry with a new node's data, possibly
+// overwriting an old position. The update is not atomic and can be racy.
+void TTEntry::save(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev) {
- assert(msb((mbSize << 20) / sizeof(TTEntry)) < 32);
+ // Preserve any existing move for the same position
+ if (m || uint16_t(k) != key16)
+ move16 = uint16_t(m);
- uint32_t size = ClusterSize << msb((mbSize << 20) / sizeof(TTEntry[ClusterSize]));
+ // Overwrite less valuable entries (cheapest checks first)
+ if (b == BOUND_EXACT || uint16_t(k) != key16 || d - DEPTH_OFFSET + 2 * pv > depth8 - 4)
+ {
+ assert(d > DEPTH_OFFSET);
+ assert(d < 256 + DEPTH_OFFSET);
- if (hashMask == size - ClusterSize)
- return;
+ key16 = uint16_t(k);
+ depth8 = uint8_t(d - DEPTH_OFFSET);
+ genBound8 = uint8_t(TT.generation8 | uint8_t(pv) << 2 | b);
+ value16 = int16_t(v);
+ eval16 = int16_t(ev);
+ }
+}
- hashMask = size - ClusterSize;
- free(mem);
- mem = calloc(size * sizeof(TTEntry) + CACHE_LINE_SIZE - 1, 1);
- if (!mem)
- {
- std::cerr << "Failed to allocate " << mbSize
- << "MB for transposition table." << std::endl;
- exit(EXIT_FAILURE);
- }
+// Sets the size of the transposition table,
+// measured in megabytes. Transposition table consists of a power of 2 number
+// of clusters and each cluster consists of ClusterSize number of TTEntry.
+void TranspositionTable::resize(size_t mbSize) {
- table = (TTEntry*)((uintptr_t(mem) + CACHE_LINE_SIZE - 1) & ~(CACHE_LINE_SIZE - 1));
-}
+ Threads.main()->wait_for_search_finished();
+ aligned_large_pages_free(table);
-/// TranspositionTable::clear() overwrites the entire transposition table
-/// with zeroes. It is called whenever the table is resized, or when the
-/// user asks the program to clear the table (from the UCI interface).
+ clusterCount = mbSize * 1024 * 1024 / sizeof(Cluster);
-void TranspositionTable::clear() {
+ table = static_cast<Cluster*>(aligned_large_pages_alloc(clusterCount * sizeof(Cluster)));
+ if (!table)
+ {
+ std::cerr << "Failed to allocate " << mbSize << "MB for transposition table." << std::endl;
+ exit(EXIT_FAILURE);
+ }
- std::memset(table, 0, (hashMask + ClusterSize) * sizeof(TTEntry));
+ clear();
}
-/// TranspositionTable::probe() looks up the current position in the
-/// transposition table. Returns a pointer to the TTEntry or NULL if
-/// position is not found.
-
-const TTEntry* TranspositionTable::probe(const Key key) const {
+// Initializes the entire transposition table to zero,
+// in a multi-threaded way.
+void TranspositionTable::clear() {
- TTEntry* tte = first_entry(key);
- uint32_t key32 = key >> 32;
+ std::vector<std::thread> threads;
- for (unsigned i = 0; i < ClusterSize; ++i, ++tte)
- if (tte->key32 == key32)
- {
- tte->generation8 = generation; // Refresh
- return tte;
- }
+ for (size_t idx = 0; idx < size_t(Options["Threads"]); ++idx)
+ {
+ threads.emplace_back([this, idx]() {
+ // Thread binding gives faster search on systems with a first-touch policy
+ if (Options["Threads"] > 8)
+ WinProcGroup::bindThisThread(idx);
- return NULL;
-}
+ // Each thread will zero its part of the hash table
+ const size_t stride = size_t(clusterCount / Options["Threads"]),
+ start = size_t(stride * idx),
+ len =
+ idx != size_t(Options["Threads"]) - 1 ? stride : clusterCount - start;
+ std::memset(&table[start], 0, len * sizeof(Cluster));
+ });
+ }
-/// TranspositionTable::store() writes a new entry containing position key and
-/// valuable information of current position. The lowest order bits of position
-/// key are used to decide in which cluster the position will be placed.
-/// When a new entry is written and there are no empty entries available in the
-/// cluster, it replaces the least valuable of the entries. A TTEntry t1 is considered
-/// to be more valuable than a TTEntry t2 if t1 is from the current search and t2
-/// is from a previous search, or if the depth of t1 is bigger than the depth of t2.
+ for (std::thread& th : threads)
+ th.join();
+}
-void TranspositionTable::store(const Key key, Value v, Bound b, Depth d, Move m, Value statV) {
- TTEntry *tte, *replace;
- uint32_t key32 = key >> 32; // Use the high 32 bits as key inside the cluster
+// Looks up the current position in the transposition
+// table. It returns true and a pointer to the TTEntry if the position is found.
+// Otherwise, it returns false and a pointer to an empty or least valuable TTEntry
+// to be replaced later. The replace value of an entry is calculated as its depth
+// minus 8 times its relative age. TTEntry t1 is considered more valuable than
+// TTEntry t2 if its replace value is greater than that of t2.
+TTEntry* TranspositionTable::probe(const Key key, bool& found) const {
+
+ TTEntry* const tte = first_entry(key);
+ const uint16_t key16 = uint16_t(key); // Use the low 16 bits as key inside the cluster
+
+ for (int i = 0; i < ClusterSize; ++i)
+ if (tte[i].key16 == key16 || !tte[i].depth8)
+ {
+ tte[i].genBound8 =
+ uint8_t(generation8 | (tte[i].genBound8 & (GENERATION_DELTA - 1))); // Refresh
+
+ return found = bool(tte[i].depth8), &tte[i];
+ }
+
+ // Find an entry to be replaced according to the replacement strategy
+ TTEntry* replace = tte;
+ for (int i = 1; i < ClusterSize; ++i)
+ // Due to our packed storage format for generation and its cyclic
+ // nature we add GENERATION_CYCLE (256 is the modulus, plus what
+ // is needed to keep the unrelated lowest n bits from affecting
+ // the result) to calculate the entry age correctly even after
+ // generation8 overflows into the next cycle.
+ if (replace->depth8
+ - ((GENERATION_CYCLE + generation8 - replace->genBound8) & GENERATION_MASK)
+ > tte[i].depth8
+ - ((GENERATION_CYCLE + generation8 - tte[i].genBound8) & GENERATION_MASK))
+ replace = &tte[i];
+
+ return found = false, replace;
+}
- tte = replace = first_entry(key);
- for (unsigned i = 0; i < ClusterSize; ++i, ++tte)
- {
- if (!tte->key32 || tte->key32 == key32) // Empty or overwrite old
- {
- if (!m)
- m = tte->move(); // Preserve any existing ttMove
+// Returns an approximation of the hashtable
+// occupation during a search. The hash is x permill full, as per UCI protocol.
- replace = tte;
- break;
- }
+int TranspositionTable::hashfull() const {
- // Implement replace strategy
- if ( (replace->generation8 == generation) * 2
- - (tte->generation8 == generation || tte->bound() == BOUND_EXACT) * 2
- + (tte->depth16 < replace->depth16) > 0)
- replace = tte;
- }
+ int cnt = 0;
+ for (int i = 0; i < 1000; ++i)
+ for (int j = 0; j < ClusterSize; ++j)
+ cnt += table[i].entry[j].depth8
+ && (table[i].entry[j].genBound8 & GENERATION_MASK) == generation8;
- replace->save(key32, v, b, d, m, generation, statV);
+ return cnt / ClusterSize;
}
+
+} // namespace Stockfish