/*
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) 2015-2017 Marco Costalba, Joona Kiiski, Gary Linscott, 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 <algorithm> // For std::count
#include <cassert>
+#include <algorithm> // For std::count
#include "movegen.h"
#include "search.h"
#include "thread.h"
#include "uci.h"
#include "syzygy/tbprobe.h"
+#include "tt.h"
+
+namespace Stockfish {
ThreadPool Threads; // Global object
-/// Thread constructor launches the thread and then waits until it goes to sleep
-/// in idle_loop().
-Thread::Thread() {
+/// Thread constructor launches the thread and waits until it goes to sleep
+/// in idle_loop(). Note that 'searching' and 'exit' should be already set.
- exit = false;
- selDepth = 0;
- nodes = tbHits = 0;
- idx = Threads.size(); // Start from 0
+Thread::Thread(size_t n) : idx(n), stdThread(&Thread::idle_loop, this) {
- std::unique_lock<Mutex> lk(mutex);
- searching = true;
- nativeThread = std::thread(&Thread::idle_loop, this);
- sleepCondition.wait(lk, [&]{ return !searching; });
+ wait_for_search_finished();
}
-/// Thread destructor waits for thread termination before returning
+/// Thread destructor wakes up the thread in idle_loop() and waits
+/// for its termination. Thread should be already waiting.
Thread::~Thread() {
- mutex.lock();
+ assert(!searching);
+
exit = true;
- sleepCondition.notify_one();
- mutex.unlock();
- nativeThread.join();
+ start_searching();
+ stdThread.join();
}
-/// Thread::wait_for_search_finished() waits on sleep condition
-/// until not searching
+/// Thread::clear() reset histories, usually before a new game
-void Thread::wait_for_search_finished() {
+void Thread::clear() {
- std::unique_lock<Mutex> lk(mutex);
- sleepCondition.wait(lk, [&]{ return !searching; });
-}
-
-
-/// Thread::wait() waits on sleep condition until condition is true
+ counterMoves.fill(MOVE_NONE);
+ mainHistory.fill(0);
+ captureHistory.fill(0);
-void Thread::wait(std::atomic_bool& condition) {
-
- std::unique_lock<Mutex> lk(mutex);
- sleepCondition.wait(lk, [&]{ return bool(condition); });
+ for (bool inCheck : { false, true })
+ for (StatsType c : { NoCaptures, Captures })
+ for (auto& to : continuationHistory[inCheck][c])
+ for (auto& h : to)
+ h->fill(-71);
}
/// Thread::start_searching() wakes up the thread that will start the search
-void Thread::start_searching(bool resume) {
+void Thread::start_searching() {
+ mutex.lock();
+ searching = true;
+ mutex.unlock(); // Unlock before notifying saves a few CPU-cycles
+ cv.notify_one(); // Wake up the thread in idle_loop()
+}
+
- std::unique_lock<Mutex> lk(mutex);
+/// Thread::wait_for_search_finished() blocks on the condition variable
+/// until the thread has finished searching.
- if (!resume)
- searching = true;
+void Thread::wait_for_search_finished() {
- sleepCondition.notify_one();
+ std::unique_lock<std::mutex> lk(mutex);
+ cv.wait(lk, [&]{ return !searching; });
}
-/// Thread::idle_loop() is where the thread is parked when it has no work to do
+/// Thread::idle_loop() is where the thread is parked, blocked on the
+/// condition variable, when it has no work to do.
void Thread::idle_loop() {
- WinProcGroup::bindThisThread(idx);
+ // If OS already scheduled us on a different group than 0 then don't overwrite
+ // the choice, eventually we are one of many one-threaded processes running on
+ // some Windows NUMA hardware, for instance in fishtest. To make it simple,
+ // just check if running threads are below a threshold, in this case all this
+ // NUMA machinery is not needed.
+ if (Options["Threads"] > 8)
+ WinProcGroup::bindThisThread(idx);
- while (!exit)
+ while (true)
{
- std::unique_lock<Mutex> lk(mutex);
-
+ std::unique_lock<std::mutex> lk(mutex);
searching = false;
+ cv.notify_one(); // Wake up anyone waiting for search finished
+ cv.wait(lk, [&]{ return searching; });
- while (!searching && !exit)
- {
- sleepCondition.notify_one(); // Wake up any waiting thread
- sleepCondition.wait(lk);
- }
+ if (exit)
+ return;
lk.unlock();
- if (!exit)
- search();
+ search();
}
}
+/// ThreadPool::set() creates/destroys threads to match the requested number.
+/// Created and launched threads will immediately go to sleep in idle_loop.
+/// Upon resizing, threads are recreated to allow for binding if necessary.
-/// ThreadPool::init() creates and launches requested threads that will go
-/// immediately to sleep. We cannot use a constructor because Threads is a
-/// static object and we need a fully initialized engine at this point due to
-/// allocation of Endgames in the Thread constructor.
-
-void ThreadPool::init() {
-
- push_back(new MainThread());
- read_uci_options();
-}
-
-
-/// ThreadPool::exit() terminates threads before the program exits. Cannot be
-/// done in destructor because threads must be terminated before deleting any
-/// static objects while still in main().
-
-void ThreadPool::exit() {
-
- while (size())
- delete back(), pop_back();
-}
+void ThreadPool::set(size_t requested) {
+ if (threads.size() > 0) // destroy any existing thread(s)
+ {
+ main()->wait_for_search_finished();
-/// ThreadPool::read_uci_options() updates internal threads parameters from the
-/// corresponding UCI options and creates/destroys threads to match requested
-/// number. Thread objects are dynamically allocated.
-
-void ThreadPool::read_uci_options() {
+ while (threads.size() > 0)
+ delete threads.back(), threads.pop_back();
+ }
- size_t requested = Options["Threads"];
+ if (requested > 0) // create new thread(s)
+ {
+ threads.push_back(new MainThread(0));
- assert(requested > 0);
+ while (threads.size() < requested)
+ threads.push_back(new Thread(threads.size()));
+ clear();
- while (size() < requested)
- push_back(new Thread());
+ // Reallocate the hash with the new threadpool size
+ TT.resize(size_t(Options["Hash"]));
- while (size() > requested)
- delete back(), pop_back();
+ // Init thread number dependent search params.
+ Search::init();
+ }
}
-/// ThreadPool::nodes_searched() returns the number of nodes searched
-
-uint64_t ThreadPool::nodes_searched() const {
+/// ThreadPool::clear() sets threadPool data to initial values
- uint64_t nodes = 0;
- for (Thread* th : *this)
- nodes += th->nodes.load(std::memory_order_relaxed);
- return nodes;
-}
-
-
-/// ThreadPool::tb_hits() returns the number of TB hits
+void ThreadPool::clear() {
-uint64_t ThreadPool::tb_hits() const {
+ for (Thread* th : threads)
+ th->clear();
- uint64_t hits = 0;
- for (Thread* th : *this)
- hits += th->tbHits.load(std::memory_order_relaxed);
- return hits;
+ main()->callsCnt = 0;
+ main()->bestPreviousScore = VALUE_INFINITE;
+ main()->bestPreviousAverageScore = VALUE_INFINITE;
+ main()->previousTimeReduction = 1.0;
}
-/// ThreadPool::start_thinking() wakes up the main thread sleeping in idle_loop()
-/// and starts a new search, then returns immediately.
+/// ThreadPool::start_thinking() wakes up main thread waiting in idle_loop() and
+/// returns immediately. Main thread will wake up other threads and start the search.
void ThreadPool::start_thinking(Position& pos, StateListPtr& states,
- const Search::LimitsType& limits) {
+ const Search::LimitsType& limits, bool ponderMode) {
main()->wait_for_search_finished();
- stopOnPonderhit = stop = false;
+ main()->stopOnPonderhit = stop = false;
+ increaseDepth = true;
+ main()->ponder = ponderMode;
Search::Limits = limits;
Search::RootMoves rootMoves;
for (const auto& m : MoveList<LEGAL>(pos))
if ( limits.searchmoves.empty()
|| std::count(limits.searchmoves.begin(), limits.searchmoves.end(), m))
- rootMoves.push_back(Search::RootMove(m));
+ rootMoves.emplace_back(m);
if (!rootMoves.empty())
- Tablebases::filter_root_moves(pos, rootMoves);
+ Tablebases::rank_root_moves(pos, rootMoves);
// After ownership transfer 'states' becomes empty, so if we stop the search
- // and call 'go' again without setting a new position states.get() == NULL.
+ // and call 'go' again without setting a new position states.get() == nullptr.
assert(states.get() || setupStates.get());
if (states.get())
setupStates = std::move(states); // Ownership transfer, states is now empty
- StateInfo tmp = setupStates->back();
-
- for (Thread* th : Threads)
+ // We use Position::set() to set root position across threads. But there are
+ // some StateInfo fields (previous, pliesFromNull, capturedPiece) that cannot
+ // be deduced from a fen string, so set() clears them and they are set from
+ // setupStates->back() later. The rootState is per thread, earlier states are shared
+ // since they are read-only.
+ for (Thread* th : threads)
{
- th->nodes = 0;
- th->tbHits = 0;
- th->rootDepth = DEPTH_ZERO;
+ th->nodes = th->tbHits = th->nmpMinPly = th->bestMoveChanges = 0;
+ th->rootDepth = th->completedDepth = 0;
th->rootMoves = rootMoves;
- th->rootPos.set(pos.fen(), pos.is_chess960(), &setupStates->back(), th);
+ th->rootPos.set(pos.fen(), pos.is_chess960(), &th->rootState, th);
+ th->rootState = setupStates->back();
}
- setupStates->back() = tmp; // Restore st->previous, cleared by Position::set()
-
main()->start_searching();
}
+
+Thread* ThreadPool::get_best_thread() const {
+
+ Thread* bestThread = threads.front();
+ std::map<Move, int64_t> votes;
+ Value minScore = VALUE_NONE;
+
+ // Find minimum score of all threads
+ for (Thread* th: threads)
+ minScore = std::min(minScore, th->rootMoves[0].score);
+
+ // Vote according to score and depth, and select the best thread
+ auto thread_value = [minScore](Thread* th) {
+ return (th->rootMoves[0].score - minScore + 14) * int(th->completedDepth);
+ };
+
+ for (Thread* th : threads)
+ votes[th->rootMoves[0].pv[0]] += thread_value(th);
+
+ for (Thread* th : threads)
+ if (abs(bestThread->rootMoves[0].score) >= VALUE_TB_WIN_IN_MAX_PLY)
+ {
+ // Make sure we pick the shortest mate / TB conversion or stave off mate the longest
+ if (th->rootMoves[0].score > bestThread->rootMoves[0].score)
+ bestThread = th;
+ }
+ else if ( th->rootMoves[0].score >= VALUE_TB_WIN_IN_MAX_PLY
+ || ( th->rootMoves[0].score > VALUE_TB_LOSS_IN_MAX_PLY
+ && ( votes[th->rootMoves[0].pv[0]] > votes[bestThread->rootMoves[0].pv[0]]
+ || ( votes[th->rootMoves[0].pv[0]] == votes[bestThread->rootMoves[0].pv[0]]
+ && thread_value(th) * int(th->rootMoves[0].pv.size() > 2)
+ > thread_value(bestThread) * int(bestThread->rootMoves[0].pv.size() > 2)))))
+ bestThread = th;
+
+ return bestThread;
+}
+
+
+/// Start non-main threads
+
+void ThreadPool::start_searching() {
+
+ for (Thread* th : threads)
+ if (th != threads.front())
+ th->start_searching();
+}
+
+
+/// Wait for non-main threads
+
+void ThreadPool::wait_for_search_finished() const {
+
+ for (Thread* th : threads)
+ if (th != threads.front())
+ th->wait_for_search_finished();
+}
+
+} // namespace Stockfish