Merge remote-tracking branch 'upstream/master'

[stockfish] / src / thread.h

diff --git a/src/thread.h b/src/thread.h
index ad40a34146bbd8e8b8fe0f80be4dcc23b5dd7d44..43e2e1423ced3cadad508f5e1b8bc2b5bfc55978 100644 (file)
--- a/src/thread.h
+++ b/src/thread.h
@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1

-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2024 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
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by


@@ -17,127 +16,163 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 

-#if !defined(THREAD_H_INCLUDED)
+#ifndef THREAD_H_INCLUDED

 #define THREAD_H_INCLUDED
 
 #define THREAD_H_INCLUDED
 

+#include <atomic>
+#include <condition_variable>
+#include <cstddef>
+#include <cstdint>
+#include <functional>
+#include <memory>
+#include <mutex>

 #include <vector>
 
 #include <vector>
 

-#include "material.h"
-#include "movepick.h"
-#include "pawns.h"
+#include "numa.h"

 #include "position.h"
 #include "search.h"
 #include "position.h"
 #include "search.h"

-
-const int MAX_THREADS = 32;
-const int MAX_SPLITPOINTS_PER_THREAD = 8;
-
-class Thread;
-
-struct SplitPoint {
-
-  // Const data after split point has been setup
-  const Position* pos;
-  const Search::Stack* ss;
-  Depth depth;
-  Value beta;
-  int nodeType;
-  Thread* master;
-  Move threatMove;
-
-  // Const pointers to shared data
-  MovePicker* mp;
-  SplitPoint* parent;
-
-  // Shared data
-  Lock lock;
-  volatile uint64_t slavesMask;
-  volatile int64_t nodes;
-  volatile Value alpha;
-  volatile Value bestValue;
-  volatile Move bestMove;
-  volatile int moveCount;
-  volatile bool cutoff;
+#include "thread_win32_osx.h"
+
+namespace Stockfish {
+
+
+class OptionsMap;
+using Value = int;
+
+// Sometimes we don't want to actually bind the threads, but the recipient still
+// needs to think it runs on *some* NUMA node, such that it can access structures
+// that rely on NUMA node knowledge. This class encapsulates this optional process
+// such that the recipient does not need to know whether the binding happened or not.
+class OptionalThreadToNumaNodeBinder {
+   public:
+    OptionalThreadToNumaNodeBinder(NumaIndex n) :
+        numaConfig(nullptr),
+        numaId(n) {}
+
+    OptionalThreadToNumaNodeBinder(const NumaConfig& cfg, NumaIndex n) :
+        numaConfig(&cfg),
+        numaId(n) {}
+
+    NumaReplicatedAccessToken operator()() const {
+        if (numaConfig != nullptr)
+            return numaConfig->bind_current_thread_to_numa_node(numaId);
+        else
+            return NumaReplicatedAccessToken(numaId);
+    }
+
+   private:
+    const NumaConfig* numaConfig;
+    NumaIndex         numaId;

 };
 
 };
 

-
-/// Thread struct keeps together all the thread related stuff like locks, state
-/// and especially split points. We also use per-thread pawn and material hash
-/// tables so that once we get a pointer to an entry its life time is unlimited
-/// and we don't have to care about someone changing the entry under our feet.
-
+// Abstraction of a thread. It contains a pointer to the worker and a native thread.
+// After construction, the native thread is started with idle_loop()
+// waiting for a signal to start searching.
+// When the signal is received, the thread starts searching and when
+// the search is finished, it goes back to idle_loop() waiting for a new signal.

 class Thread {
 class Thread {

-
-  typedef void (Thread::* Fn) (); // Pointer to member function
-
-public:
-  Thread(Fn fn);
- ~Thread();
-
-  void wake_up();
-  bool cutoff_occurred() const;
-  bool is_available_to(Thread* master) const;
-  void idle_loop();
-  void main_loop();
-  void timer_loop();
-  void wait_for_stop_or_ponderhit();
-
-  SplitPoint splitPoints[MAX_SPLITPOINTS_PER_THREAD];
-  MaterialTable materialTable;
-  PawnTable pawnTable;
-  size_t idx;
-  int maxPly;
-  Lock sleepLock;
-  WaitCondition sleepCond;
-  NativeHandle handle;
-  Fn start_fn;
-  SplitPoint* volatile curSplitPoint;
-  volatile int splitPointsCnt;
-  volatile bool is_searching;
-  volatile bool do_sleep;
-  volatile bool do_exit;
+   public:
+    Thread(Search::SharedState&,
+           std::unique_ptr<Search::ISearchManager>,
+           size_t,
+           OptionalThreadToNumaNodeBinder);
+    virtual ~Thread();
+
+    void idle_loop();
+    void start_searching();
+    void clear_worker();
+    void run_custom_job(std::function<void()> f);
+
+    void ensure_network_replicated();
+
+    // Thread has been slightly altered to allow running custom jobs, so
+    // this name is no longer correct. However, this class (and ThreadPool)
+    // require further work to make them properly generic while maintaining
+    // appropriate specificity regarding search, from the point of view of an
+    // outside user, so renaming of this function is left for whenever that happens.
+    void   wait_for_search_finished();
+    size_t id() const { return idx; }
+
+    std::unique_ptr<Search::Worker> worker;
+    std::function<void()>           jobFunc;
+
+   private:
+    std::mutex                mutex;
+    std::condition_variable   cv;
+    size_t                    idx, nthreads;
+    bool                      exit = false, searching = true;  // Set before starting std::thread
+    NativeThread              stdThread;
+    NumaReplicatedAccessToken numaAccessToken;

 };
 
 
 };
 
 

-/// ThreadPool class handles all the threads related stuff like init, starting,
-/// parking and, the most important, launching a slave thread at a split point.
-/// All the access to shared thread data is done through this class.
-
+// ThreadPool struct handles all the threads-related stuff like init, starting,
+// parking and, most importantly, launching a thread. All the access to threads
+// is done through this class.

 class ThreadPool {
 class ThreadPool {

-
-public:
-  void init(); // No c'tor, Threads object is global and engine shall be fully initialized
-  ~ThreadPool();
-
-  Thread& operator[](size_t id) { return *threads[id]; }
-  bool use_sleeping_threads() const { return useSleepingThreads; }
-  int min_split_depth() const { return minimumSplitDepth; }
-  size_t size() const { return threads.size(); }
-  Thread* main_thread() { return threads[0]; }
-
-  void wake_up() const;
-  void sleep() const;
-  void read_uci_options();
-  bool available_slave_exists(Thread* master) const;
-  void set_timer(int msec);
-  void wait_for_search_finished();
-  void start_searching(const Position& pos, const Search::LimitsType& limits,
-                       const std::vector<Move>& searchMoves);
-
-  template <bool Fake>
-  Value split(Position& pos, Search::Stack* ss, Value alpha, Value beta, Value bestValue, Move* bestMove,
-              Depth depth, Move threatMove, int moveCount, MovePicker* mp, int nodeType);
-private:
-  friend class Thread;
-
-  std::vector<Thread*> threads;
-  Thread* timer;
-  Lock splitLock;
-  WaitCondition sleepCond;
-  Depth minimumSplitDepth;
-  int maxThreadsPerSplitPoint;
-  bool useSleepingThreads;
+   public:
+    ThreadPool() {}
+
+    ~ThreadPool() {
+        // destroy any existing thread(s)
+        if (threads.size() > 0)
+        {
+            main_thread()->wait_for_search_finished();
+
+            threads.clear();
+        }
+    }
+
+    ThreadPool(const ThreadPool&) = delete;
+    ThreadPool(ThreadPool&&)      = delete;
+
+    ThreadPool& operator=(const ThreadPool&) = delete;
+    ThreadPool& operator=(ThreadPool&&)      = delete;
+
+    void   start_thinking(const OptionsMap&, Position&, StateListPtr&, Search::LimitsType);
+    void   run_on_thread(size_t threadId, std::function<void()> f);
+    void   wait_on_thread(size_t threadId);
+    size_t num_threads() const;
+    void   clear();
+    void   set(const NumaConfig& numaConfig,
+               Search::SharedState,
+               const Search::SearchManager::UpdateContext&);
+
+    Search::SearchManager* main_manager();
+    Thread*                main_thread() const { return threads.front().get(); }
+    uint64_t               nodes_searched() const;
+    uint64_t               tb_hits() const;
+    Thread*                get_best_thread() const;
+    void                   start_searching();
+    void                   wait_for_search_finished() const;
+
+    std::vector<size_t> get_bound_thread_count_by_numa_node() const;
+
+    void ensure_network_replicated();
+
+    std::atomic_bool stop, abortedSearch, increaseDepth;
+
+    auto cbegin() const noexcept { return threads.cbegin(); }
+    auto begin() noexcept { return threads.begin(); }
+    auto end() noexcept { return threads.end(); }
+    auto cend() const noexcept { return threads.cend(); }
+    auto size() const noexcept { return threads.size(); }
+    auto empty() const noexcept { return threads.empty(); }
+
+   private:
+    StateListPtr                         setupStates;
+    std::vector<std::unique_ptr<Thread>> threads;
+    std::vector<NumaIndex>               boundThreadToNumaNode;
+
+    uint64_t accumulate(std::atomic<uint64_t> Search::Worker::*member) const {
+
+        uint64_t sum = 0;
+        for (auto&& th : threads)
+            sum += (th->worker.get()->*member).load(std::memory_order_relaxed);
+        return sum;
+    }

 };
 
 };
 

-extern ThreadPool Threads;
+}  // namespace Stockfish

 
 

-#endif // !defined(THREAD_H_INCLUDED)
+#endif  // #ifndef THREAD_H_INCLUDED