X-Git-Url: https://git.sesse.net/?p=stockfish;a=blobdiff_plain;f=src%2Ftt.cpp;h=cb5af5c85905620a1537d06a7339ce6bca17176f;hp=26791809f8890bb6aed2e492838b9a9473b2aaf2;hb=573f0e364ff4c1e5928be2ca947f65c5d4e177d5;hpb=e7cfa5d020efb5a0ad2521afc7b886f3b2d3e6b3 diff --git a/src/tt.cpp b/src/tt.cpp index 26791809..cb5af5c8 100644 --- a/src/tt.cpp +++ b/src/tt.cpp @@ -1,8 +1,6 @@ /* Stockfish, a UCI chess playing engine derived from Glaurung 2.1 - Copyright (C) 2004-2008 Tord Romstad (Glaurung author) - Copyright (C) 2008-2015 Marco Costalba, Joona Kiiski, Tord Romstad - Copyright (C) 2015-2018 Marco Costalba, Joona Kiiski, Gary Linscott, Tord Romstad + Copyright (C) 2004-2021 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 @@ -24,30 +22,34 @@ #include "bitboard.h" #include "misc.h" +#include "thread.h" #include "tt.h" #include "uci.h" TranspositionTable TT; // Our global transposition table -/// TTEntry::save saves a TTEntry -void TTEntry::save(Key k, Value v, Bound b, Depth d, Move m, Value ev) { +/// TTEntry::save() populates the TTEntry with a new node's data, possibly +/// overwriting an old position. Update is not atomic and can be racy. - assert(d / ONE_PLY * ONE_PLY == d); +void TTEntry::save(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev) { // Preserve any existing move for the same position - if (m || (k >> 48) != key16) + if (m || (uint16_t)k != key16) move16 = (uint16_t)m; - // Overwrite less valuable entries - if ( (k >> 48) != key16 - || d / ONE_PLY > depth8 - 4 - || b == BOUND_EXACT) + // Overwrite less valuable entries (cheapest checks first) + if (b == BOUND_EXACT + || (uint16_t)k != key16 + || d - DEPTH_OFFSET > depth8 - 4) { - key16 = (uint16_t)(k >> 48); + assert(d > DEPTH_OFFSET); + assert(d < 256 + DEPTH_OFFSET); + + 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; - genBound8 = (uint8_t)(TT.generation8 | b); - depth8 = (int8_t)(d / ONE_PLY); } } @@ -58,19 +60,20 @@ void TTEntry::save(Key k, Value v, Bound b, Depth d, Move m, Value ev) { void TranspositionTable::resize(size_t mbSize) { - clusterCount = mbSize * 1024 * 1024 / sizeof(Cluster); + Threads.main()->wait_for_search_finished(); - free(mem); - mem = malloc(clusterCount * sizeof(Cluster) + CacheLineSize - 1); + aligned_large_pages_free(table); + + clusterCount = mbSize * 1024 * 1024 / sizeof(Cluster); - if (!mem) + table = static_cast(aligned_large_pages_alloc(clusterCount * sizeof(Cluster))); + if (!table) { std::cerr << "Failed to allocate " << mbSize << "MB for transposition table." << std::endl; exit(EXIT_FAILURE); } - table = (Cluster*)((uintptr_t(mem) + CacheLineSize - 1) & ~(CacheLineSize - 1)); clear(); } @@ -82,28 +85,29 @@ void TranspositionTable::clear() { std::vector threads; - for (size_t idx = 0; idx < Options["Threads"]; idx++) + for (size_t idx = 0; idx < Options["Threads"]; ++idx) { - threads.push_back(std::thread([this, idx]() { + threads.emplace_back([this, idx]() { // Thread binding gives faster search on systems with a first-touch policy - if (Options["Threads"] >= 8) + if (Options["Threads"] > 8) WinProcGroup::bindThisThread(idx); // Each thread will zero its part of the hash table - const size_t stride = clusterCount / Options["Threads"], - start = stride * idx, + const size_t stride = size_t(clusterCount / Options["Threads"]), + start = size_t(stride * idx), len = idx != Options["Threads"] - 1 ? stride : clusterCount - start; std::memset(&table[start], 0, len * sizeof(Cluster)); - })); + }); } - for (std::thread& th: threads) + for (std::thread& th : threads) th.join(); } + /// TranspositionTable::probe() 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 @@ -114,26 +118,26 @@ void TranspositionTable::clear() { TTEntry* TranspositionTable::probe(const Key key, bool& found) const { TTEntry* const tte = first_entry(key); - const uint16_t key16 = key >> 48; // Use the high 16 bits as key inside the cluster + 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 || tte[i].key16 == key16) + if (tte[i].key16 == key16 || !tte[i].depth8) { - if ((tte[i].genBound8 & 0xFC) != generation8 && tte[i].key16) - tte[i].genBound8 = uint8_t(generation8 | tte[i].bound()); // Refresh + tte[i].genBound8 = uint8_t(generation8 | (tte[i].genBound8 & (GENERATION_DELTA - 1))); // Refresh - return found = (bool)tte[i].key16, &tte[i]; + 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 259 (256 is the modulus plus 3 to keep the lowest - // two bound bits from affecting the result) to calculate the entry - // age correctly even after generation8 overflows into the next cycle. - if ( replace->depth8 - ((259 + generation8 - replace->genBound8) & 0xFC) * 2 - > tte[i].depth8 - ((259 + generation8 - tte[i].genBound8) & 0xFC) * 2) + // 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; @@ -146,12 +150,9 @@ TTEntry* TranspositionTable::probe(const Key key, bool& found) const { int TranspositionTable::hashfull() const { int cnt = 0; - for (int i = 0; i < 1000 / ClusterSize; i++) - { - const TTEntry* tte = &table[i].entry[0]; - for (int j = 0; j < ClusterSize; j++) - if ((tte[j].genBound8 & 0xFC) == generation8) - cnt++; - } - return cnt; + 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; + + return cnt / ClusterSize; }