X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=src%2Fsyzygy%2Ftbprobe.cpp;h=d1b32d242c9fd6adf2a2ae91a68ccb7bf6a271b0;hb=3c0e86a91e48baea273306e45fb6cf13a59373cf;hp=f1fd695c34a620c3102d048da20b45d92c1b94d6;hpb=f83cb95740de019db6ff5567d6f84f218b18cd9e;p=stockfish diff --git a/src/syzygy/tbprobe.cpp b/src/syzygy/tbprobe.cpp index f1fd695c..d1b32d24 100644 --- a/src/syzygy/tbprobe.cpp +++ b/src/syzygy/tbprobe.cpp @@ -1,7 +1,6 @@ /* Stockfish, a UCI chess playing engine derived from Glaurung 2.1 - Copyright (c) 2013 Ronald de Man - Copyright (C) 2016-2020 Marco Costalba, Lucas Braesch + 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 @@ -17,32 +16,38 @@ along with this program. If not, see . */ +#include "tbprobe.h" + +#include #include #include +#include #include -#include // For std::memset and std::memcpy +#include +#include #include #include +#include #include -#include +#include #include +#include #include -#include +#include +#include #include "../bitboard.h" +#include "../misc.h" #include "../movegen.h" #include "../position.h" #include "../search.h" #include "../types.h" #include "../uci.h" -#include "tbprobe.h" - #ifndef _WIN32 #include -#include #include -#include +#include #else #define WIN32_LEAN_AND_MEAN #ifndef NOMINMAX @@ -51,16 +56,19 @@ #include #endif -using namespace Tablebases; +using namespace Stockfish::Tablebases; + +int Stockfish::Tablebases::MaxCardinality; -int Tablebases::MaxCardinality; +namespace Stockfish { namespace { constexpr int TBPIECES = 7; // Max number of supported pieces +constexpr int MAX_DTZ = 1 << 18; // Max DTZ supported, large enough to deal with the syzygy TB limit. enum { BigEndian, LittleEndian }; -enum TBType { KEY, WDL, DTZ }; // Used as template parameter +enum TBType { WDL, DTZ }; // Used as template parameter // Each table has a set of flags: all of them refer to DTZ tables, the last one to WDL tables enum TBFlag { STM = 1, Mapped = 2, WinPlies = 4, LossPlies = 8, Wide = 16, SingleValue = 128 }; @@ -68,7 +76,7 @@ enum TBFlag { STM = 1, Mapped = 2, WinPlies = 4, LossPlies = 8, Wide = 16, Singl inline WDLScore operator-(WDLScore d) { return WDLScore(-int(d)); } inline Square operator^(Square s, int i) { return Square(int(s) ^ i); } -const std::string PieceToChar = " PNBRQK pnbrqk"; +constexpr std::string_view PieceToChar = " PNBRQK pnbrqk"; int MapPawns[SQUARE_NB]; int MapB1H1H7[SQUARE_NB]; @@ -104,9 +112,6 @@ template<> inline void swap_endian(uint8_t&) {} template T number(void* addr) { - static const union { uint32_t i; char c[4]; } Le = { 0x01020304 }; - static const bool IsLittleEndian = (Le.c[0] == 4); - T v; if ((uintptr_t)addr & (alignof(T) - 1)) // Unaligned pointer (very rare) @@ -142,7 +147,7 @@ struct SparseEntry { static_assert(sizeof(SparseEntry) == 6, "SparseEntry must be 6 bytes"); -typedef uint16_t Sym; // Huffman symbol +using Sym = uint16_t; // Huffman symbol struct LR { enum Side { Left, Right }; @@ -191,7 +196,8 @@ public: std::stringstream ss(Paths); std::string path; - while (std::getline(ss, path, SepChar)) { + while (std::getline(ss, path, SepChar)) + { fname = path + "/" + f; std::ifstream::open(fname); if (is_open()) @@ -199,13 +205,10 @@ public: } } - // Memory map the file and check it. File should be already open and will be - // closed after mapping. + // Memory map the file and check it. uint8_t* map(void** baseAddress, uint64_t* mapping, TBType type) { - - assert(is_open()); - - close(); // Need to re-open to get native file descriptor + if (is_open()) + close(); // Need to re-open to get native file descriptor #ifndef _WIN32 struct stat statbuf; @@ -224,7 +227,9 @@ public: *mapping = statbuf.st_size; *baseAddress = mmap(nullptr, statbuf.st_size, PROT_READ, MAP_SHARED, fd, 0); +#if defined(MADV_RANDOM) madvise(*baseAddress, statbuf.st_size, MADV_RANDOM); +#endif ::close(fd); if (*baseAddress == MAP_FAILED) @@ -234,7 +239,7 @@ public: } #else // Note FILE_FLAG_RANDOM_ACCESS is only a hint to Windows and as such may get ignored. - HANDLE fd = CreateFile(fname.c_str(), GENERIC_READ, FILE_SHARE_READ, nullptr, + HANDLE fd = CreateFileA(fname.c_str(), GENERIC_READ, FILE_SHARE_READ, nullptr, OPEN_EXISTING, FILE_FLAG_RANDOM_ACCESS, nullptr); if (fd == INVALID_HANDLE_VALUE) @@ -327,7 +332,7 @@ struct PairsData { // first access, when the corresponding file is memory mapped. template struct TBTable { - typedef typename std::conditional::type Ret; + using Ret = typename std::conditional::type; static constexpr int Sides = Type == WDL ? 2 : 1; @@ -403,7 +408,17 @@ TBTable::TBTable(const TBTable& wdl) : TBTable() { // at init time, accessed at probe time. class TBTables { - typedef std::tuple*, TBTable*> Entry; + struct Entry + { + Key key; + TBTable* wdl; + TBTable* dtz; + + template + TBTable* get() const { + return (TBTable*)(Type == WDL ? (void*)wdl : (void*)dtz); + } + }; static constexpr int Size = 1 << 12; // 4K table, indexed by key's 12 lsb static constexpr int Overflow = 1; // Number of elements allowed to map to the last bucket @@ -415,12 +430,12 @@ class TBTables { void insert(Key key, TBTable* wdl, TBTable* dtz) { uint32_t homeBucket = (uint32_t)key & (Size - 1); - Entry entry = std::make_tuple(key, wdl, dtz); + Entry entry{ key, wdl, dtz }; // Ensure last element is empty to avoid overflow when looking up for (uint32_t bucket = homeBucket; bucket < Size + Overflow - 1; ++bucket) { - Key otherKey = std::get(hashTable[bucket]); - if (otherKey == key || !std::get(hashTable[bucket])) { + Key otherKey = hashTable[bucket].key; + if (otherKey == key || !hashTable[bucket].get()) { hashTable[bucket] = entry; return; } @@ -429,7 +444,7 @@ class TBTables { // insert here and search for a new spot for the other element instead. uint32_t otherHomeBucket = (uint32_t)otherKey & (Size - 1); if (otherHomeBucket > homeBucket) { - swap(entry, hashTable[bucket]); + std::swap(entry, hashTable[bucket]); key = otherKey; homeBucket = otherHomeBucket; } @@ -442,8 +457,8 @@ public: template TBTable* get(Key key) { for (const Entry* entry = &hashTable[(uint32_t)key & (Size - 1)]; ; ++entry) { - if (std::get(*entry) == key || !std::get(*entry)) - return std::get(*entry); + if (entry->key == key || !entry->get()) + return entry->get(); } } @@ -520,7 +535,7 @@ int decompress_pairs(PairsData* d, uint64_t idx) { // I(k) = k * d->span + d->span / 2 (1) // First step is to get the 'k' of the I(k) nearest to our idx, using definition (1) - uint32_t k = idx / d->span; + uint32_t k = uint32_t(idx / d->span); // Then we read the corresponding SparseIndex[] entry uint32_t block = number(&d->sparseIndex[k].block); @@ -554,7 +569,8 @@ int decompress_pairs(PairsData* d, uint64_t idx) { int buf64Size = 64; Sym sym; - while (true) { + while (true) + { int len = 0; // This is the symbol length - d->min_sym_len // Now get the symbol length. For any symbol s64 of length l right-padded @@ -566,7 +582,7 @@ int decompress_pairs(PairsData* d, uint64_t idx) { // All the symbols of a given length are consecutive integers (numerical // sequence property), so we can compute the offset of our symbol of // length len, stored at the beginning of buf64. - sym = (buf64 - d->base64[len]) >> (64 - len - d->minSymLen); + sym = Sym((buf64 - d->base64[len]) >> (64 - len - d->minSymLen)); // Now add the value of the lowest symbol of length len to get our symbol sym += number(&d->lowestSym[len]); @@ -592,8 +608,8 @@ int decompress_pairs(PairsData* d, uint64_t idx) { // We binary-search for our value recursively expanding into the left and // right child symbols until we reach a leaf node where symlen[sym] + 1 == 1 // that will store the value we need. - while (d->symlen[sym]) { - + while (d->symlen[sym]) + { Sym left = d->btree[sym].get(); // If a symbol contains 36 sub-symbols (d->symlen[sym] + 1 = 36) and @@ -698,7 +714,7 @@ Ret do_probe_table(const Position& pos, T* entry, WDLScore wdl, ProbeState* resu leadPawns = b = pos.pieces(color_of(pc), PAWN); do - squares[size++] = pop_lsb(&b) ^ flipSquares; + squares[size++] = pop_lsb(b) ^ flipSquares; while (b); leadPawnsCnt = size; @@ -718,7 +734,7 @@ Ret do_probe_table(const Position& pos, T* entry, WDLScore wdl, ProbeState* resu // directly map them to the correct color and square. b = pos.pieces() ^ leadPawns; do { - Square s = pop_lsb(&b); + Square s = pop_lsb(b); squares[size] = s ^ flipSquares; pieces[size++] = Piece(pos.piece_on(s) ^ flipColor); } while (b); @@ -749,7 +765,7 @@ Ret do_probe_table(const Position& pos, T* entry, WDLScore wdl, ProbeState* resu if (entry->hasPawns) { idx = LeadPawnIdx[leadPawnsCnt][squares[0]]; - std::sort(squares + 1, squares + leadPawnsCnt, pawns_comp); + std::stable_sort(squares + 1, squares + leadPawnsCnt, pawns_comp); for (int i = 1; i < leadPawnsCnt; ++i) idx += Binomial[i][MapPawns[squares[i]]]; @@ -757,7 +773,7 @@ Ret do_probe_table(const Position& pos, T* entry, WDLScore wdl, ProbeState* resu goto encode_remaining; // With pawns we have finished special treatments } - // In positions withouth pawns, we further flip the squares to ensure leading + // In positions without pawns, we further flip the squares to ensure leading // piece is below RANK_5. if (rank_of(squares[0]) > RANK_4) for (int i = 0; i < size; ++i) @@ -800,7 +816,7 @@ Ret do_probe_table(const Position& pos, T* entry, WDLScore wdl, ProbeState* resu // Rs "together" in 62 * 61 / 2 ways (we divide by 2 because rooks can be // swapped and still get the same position.) // - // In case we have at least 3 unique pieces (inlcuded kings) we encode them + // In case we have at least 3 unique pieces (included kings) we encode them // together. if (entry->hasUniquePieces) { @@ -815,7 +831,7 @@ Ret do_probe_table(const Position& pos, T* entry, WDLScore wdl, ProbeState* resu + (squares[1] - adjust1)) * 62 + squares[2] - adjust2; - // First piece is on a1-h8 diagonal, second below: map this occurence to + // First piece is on a1-h8 diagonal, second below: map this occurrence to // 6 to differentiate from the above case, rank_of() maps a1-d4 diagonal // to 0...3 and finally MapB1H1H7[] maps the b1-h1-h7 triangle to 0..27. else if (off_A1H8(squares[1])) @@ -845,12 +861,12 @@ encode_remaining: idx *= d->groupIdx[0]; Square* groupSq = squares + d->groupLen[0]; - // Encode remainig pawns then pieces according to square, in ascending order + // Encode remaining pawns then pieces according to square, in ascending order bool remainingPawns = entry->hasPawns && entry->pawnCount[1]; while (d->groupLen[++next]) { - std::sort(groupSq, groupSq + d->groupLen[next]); + std::stable_sort(groupSq, groupSq + d->groupLen[next]); uint64_t n = 0; // Map down a square if "comes later" than a square in the previous @@ -873,7 +889,7 @@ encode_remaining: // Group together pieces that will be encoded together. The general rule is that // a group contains pieces of same type and color. The exception is the leading -// group that, in case of positions withouth pawns, can be formed by 3 different +// group that, in case of positions without pawns, can be formed by 3 different // pieces (default) or by the king pair when there is not a unique piece apart // from the kings. When there are pawns, pawns are always first in pieces[]. // @@ -905,7 +921,7 @@ void set_groups(T& e, PairsData* d, int order[], File f) { // // This ensures unique encoding for the whole position. The order of the // groups is a per-table parameter and could not follow the canonical leading - // pawns/pieces -> remainig pawns -> remaining pieces. In particular the + // pawns/pieces -> remaining pawns -> remaining pieces. In particular the // first group is at order[0] position and the remaining pawns, when present, // are at order[1] position. bool pp = e.hasPawns && e.pawnCount[1]; // Pawns on both sides @@ -925,7 +941,7 @@ void set_groups(T& e, PairsData* d, int order[], File f) { d->groupIdx[1] = idx; idx *= Binomial[d->groupLen[1]][48 - d->groupLen[0]]; } - else // Remainig pieces + else // Remaining pieces { d->groupIdx[next] = idx; idx *= Binomial[d->groupLen[next]][freeSquares]; @@ -935,7 +951,7 @@ void set_groups(T& e, PairsData* d, int order[], File f) { d->groupIdx[n] = idx; } -// In Recursive Pairing each symbol represents a pair of childern symbols. So +// In Recursive Pairing each symbol represents a pair of children symbols. So // read d->btree[] symbols data and expand each one in his left and right child // symbol until reaching the leafs that represent the symbol value. uint8_t set_symlen(PairsData* d, Sym s, std::vector& visited) { @@ -974,7 +990,7 @@ uint8_t* set_sizes(PairsData* d, uint8_t* data) { d->sizeofBlock = 1ULL << *data++; d->span = 1ULL << *data++; - d->sparseIndexSize = (tbSize + d->span - 1) / d->span; // Round up + d->sparseIndexSize = size_t((tbSize + d->span - 1) / d->span); // Round up auto padding = number(data++); d->blocksNum = number(data); data += sizeof(uint32_t); d->blockLengthSize = d->blocksNum + padding; // Padded to ensure SparseIndex[] @@ -984,13 +1000,19 @@ uint8_t* set_sizes(PairsData* d, uint8_t* data) { d->lowestSym = (Sym*)data; d->base64.resize(d->maxSymLen - d->minSymLen + 1); + // See https://en.wikipedia.org/wiki/Huffman_coding // The canonical code is ordered such that longer symbols (in terms of // the number of bits of their Huffman code) have lower numeric value, // so that d->lowestSym[i] >= d->lowestSym[i+1] (when read as LittleEndian). // Starting from this we compute a base64[] table indexed by symbol length // and containing 64 bit values so that d->base64[i] >= d->base64[i+1]. - // See http://www.eecs.harvard.edu/~michaelm/E210/huffman.pdf - for (int i = d->base64.size() - 2; i >= 0; --i) { + + // Implementation note: we first cast the unsigned size_t "base64.size()" + // to a signed int "base64_size" variable and then we are able to subtract 2, + // avoiding unsigned overflow warnings. + + int base64_size = static_cast(d->base64.size()); + for (int i = base64_size - 2; i >= 0; --i) { d->base64[i] = (d->base64[i + 1] + number(&d->lowestSym[i]) - number(&d->lowestSym[i + 1])) / 2; @@ -1001,10 +1023,10 @@ uint8_t* set_sizes(PairsData* d, uint8_t* data) { // than d->base64[i+1] and given the above assert condition, we ensure that // d->base64[i] >= d->base64[i+1]. Moreover for any symbol s64 of length i // and right-padded to 64 bits holds d->base64[i-1] >= s64 >= d->base64[i]. - for (size_t i = 0; i < d->base64.size(); ++i) + for (int i = 0; i < base64_size; ++i) d->base64[i] <<= 64 - i - d->minSymLen; // Right-padding to 64 bits - data += d->base64.size() * sizeof(Sym); + data += base64_size * sizeof(Sym); d->symlen.resize(number(data)); data += sizeof(uint16_t); d->btree = (LR*)data; @@ -1012,7 +1034,7 @@ uint8_t* set_sizes(PairsData* d, uint8_t* data) { // frequent adjacent pair of symbols in the source message by a new symbol, // reevaluating the frequencies of all of the symbol pairs with respect to // the extended alphabet, and then repeating the process. - // See http://www.larsson.dogma.net/dcc99.pdf + // See https://web.archive.org/web/20201106232444/http://www.larsson.dogma.net/dcc99.pdf std::vector visited(d->symlen.size()); for (Sym sym = 0; sym < d->symlen.size(); ++sym) @@ -1130,7 +1152,7 @@ void* mapped(TBTable& e, const Position& pos) { if (e.ready.load(std::memory_order_acquire)) return e.baseAddress; // Could be nullptr if file does not exist - std::unique_lock lk(mutex); + std::scoped_lock lk(mutex); if (e.ready.load(std::memory_order_relaxed)) // Recheck under lock return e.baseAddress; @@ -1190,7 +1212,7 @@ WDLScore search(Position& pos, ProbeState* result) { auto moveList = MoveList(pos); size_t totalCount = moveList.size(), moveCount = 0; - for (const Move& move : moveList) + for (const Move move : moveList) { if ( !pos.capture(move) && (!CheckZeroingMoves || type_of(pos.moved_piece(move)) != PAWN)) @@ -1278,7 +1300,7 @@ void Tablebases::init(const std::string& paths) { for (auto s : diagonal) MapA1D1D4[s] = code++; - // MapKK[] encodes all the 461 possible legal positions of two kings where + // MapKK[] encodes all the 462 possible legal positions of two kings where // the first is in the a1-d1-d4 triangle. If the first king is on the a1-d4 // diagonal, the other one shall not to be above the a1-h8 diagonal. std::vector> bothOnDiagonal; @@ -1305,7 +1327,7 @@ void Tablebases::init(const std::string& paths) { for (auto p : bothOnDiagonal) MapKK[p.first][p.second] = code++; - // Binomial[] stores the Binomial Coefficents using Pascal rule. There + // Binomial[] stores the Binomial Coefficients using Pascal rule. There // are Binomial[k][n] ways to choose k elements from a set of n elements. Binomial[0][0] = 1; @@ -1325,7 +1347,7 @@ void Tablebases::init(const std::string& paths) { for (int leadPawnsCnt = 1; leadPawnsCnt <= 5; ++leadPawnsCnt) for (File f = FILE_A; f <= FILE_D; ++f) { - // Restart the index at every file because TB table is splitted + // Restart the index at every file because TB table is split // by file, so we can reuse the same index for different files. int idx = 0; @@ -1352,7 +1374,7 @@ void Tablebases::init(const std::string& paths) { LeadPawnsSize[leadPawnsCnt][f] = idx; } - // Add entries in TB tables if the corresponding ".rtbw" file exsists + // Add entries in TB tables if the corresponding ".rtbw" file exists for (PieceType p1 = PAWN; p1 < KING; ++p1) { TBTables.add({KING, p1, KING}); @@ -1429,7 +1451,7 @@ WDLScore Tablebases::probe_wdl(Position& pos, ProbeState* result) { // If n = 100 immediately after a capture or pawn move, then the position // is also certainly a win, and during the whole phase until the next // capture or pawn move, the inequality to be preserved is -// dtz + 50-movecounter <= 100. +// dtz + 50-move-counter <= 100. // // In short, if a move is available resulting in dtz + 50-move-counter <= 99, // then do not accept moves leading to dtz + 50-move-counter == 100. @@ -1459,7 +1481,7 @@ int Tablebases::probe_dtz(Position& pos, ProbeState* result) { StateInfo st; int minDTZ = 0xFFFF; - for (const Move& move : MoveList(pos)) + for (const Move move : MoveList(pos)) { bool zeroing = pos.capture(move) || type_of(pos.moved_piece(move)) == PAWN; @@ -1501,7 +1523,7 @@ int Tablebases::probe_dtz(Position& pos, ProbeState* result) { // A return value false indicates that not all probes were successful. bool Tablebases::root_probe(Position& pos, Search::RootMoves& rootMoves) { - ProbeState result; + ProbeState result = OK; StateInfo st; // Obtain 50-move counter for the root position @@ -1510,7 +1532,7 @@ bool Tablebases::root_probe(Position& pos, Search::RootMoves& rootMoves) { // Check whether a position was repeated since the last zeroing move. bool rep = pos.has_repeated(); - int dtz, bound = Options["Syzygy50MoveRule"] ? 900 : 1; + int dtz, bound = Options["Syzygy50MoveRule"] ? (MAX_DTZ - 100) : 1; // Probe and rank each move for (auto& m : rootMoves) @@ -1524,6 +1546,14 @@ bool Tablebases::root_probe(Position& pos, Search::RootMoves& rootMoves) { WDLScore wdl = -probe_wdl(pos, &result); dtz = dtz_before_zeroing(wdl); } + else if (pos.is_draw(1)) + { + // In case a root move leads to a draw by repetition or + // 50-move rule, we set dtz to zero. Note: since we are + // only 1 ply from the root, this must be a true 3-fold + // repetition inside the game history. + dtz = 0; + } else { // Otherwise, take dtz for the new position and correct by 1 ply @@ -1545,8 +1575,8 @@ bool Tablebases::root_probe(Position& pos, Search::RootMoves& rootMoves) { // Better moves are ranked higher. Certain wins are ranked equally. // Losing moves are ranked equally unless a 50-move draw is in sight. - int r = dtz > 0 ? (dtz + cnt50 <= 99 && !rep ? 1000 : 1000 - (dtz + cnt50)) - : dtz < 0 ? (-dtz * 2 + cnt50 < 100 ? -1000 : -1000 + (-dtz + cnt50)) + int r = dtz > 0 ? (dtz + cnt50 <= 99 && !rep ? MAX_DTZ : MAX_DTZ - (dtz + cnt50)) + : dtz < 0 ? (-dtz * 2 + cnt50 < 100 ? -MAX_DTZ : -MAX_DTZ + (-dtz + cnt50)) : 0; m.tbRank = r; @@ -1554,9 +1584,9 @@ bool Tablebases::root_probe(Position& pos, Search::RootMoves& rootMoves) { // 1 cp to cursed wins and let it grow to 49 cp as the positions gets // closer to a real win. m.tbScore = r >= bound ? VALUE_MATE - MAX_PLY - 1 - : r > 0 ? Value((std::max( 3, r - 800) * int(PawnValueEg)) / 200) + : r > 0 ? Value((std::max( 3, r - (MAX_DTZ - 200)) * int(PawnValue)) / 200) : r == 0 ? VALUE_DRAW - : r > -bound ? Value((std::min(-3, r + 800) * int(PawnValueEg)) / 200) + : r > -bound ? Value((std::min(-3, r + (MAX_DTZ - 200)) * int(PawnValue)) / 200) : -VALUE_MATE + MAX_PLY + 1; } @@ -1570,10 +1600,11 @@ bool Tablebases::root_probe(Position& pos, Search::RootMoves& rootMoves) { // A return value false indicates that not all probes were successful. bool Tablebases::root_probe_wdl(Position& pos, Search::RootMoves& rootMoves) { - static const int WDL_to_rank[] = { -1000, -899, 0, 899, 1000 }; + static const int WDL_to_rank[] = { -MAX_DTZ, -MAX_DTZ + 101, 0, MAX_DTZ - 101, MAX_DTZ }; - ProbeState result; + ProbeState result = OK; StateInfo st; + WDLScore wdl; bool rule50 = Options["Syzygy50MoveRule"]; @@ -1582,7 +1613,10 @@ bool Tablebases::root_probe_wdl(Position& pos, Search::RootMoves& rootMoves) { { pos.do_move(m.pv[0], st); - WDLScore wdl = -probe_wdl(pos, &result); + if (pos.is_draw(1)) + wdl = WDLDraw; + else + wdl = -probe_wdl(pos, &result); pos.undo_move(m.pv[0]); @@ -1599,3 +1633,5 @@ bool Tablebases::root_probe_wdl(Position& pos, Search::RootMoves& rootMoves) { return true; } + +} // namespace Stockfish