Revert thread_local stuff

Unfortunatly accessing thread local variable
is much slower than object data (see previous
patch log msg), so we have to revert to old code
to avoid speed regression.

No functional change.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

src/benchmark.cpp

@@ -107,7 +107,7 @@ void benchmark(istringstream& is) {
 
   for (size_t i = 0; i < fens.size(); i++)
   {
-      Position pos(fens[i], false);
+      Position pos(fens[i], false, NULL);
 
       cerr << "\nPosition: " << i + 1 << '/' << fens.size() << endl;
 

src/endgame.cpp

@@ -77,7 +77,7 @@ namespace {
     string fen =  sides[0] + char('0' + int(8 - code.length()))
                 + sides[1] + "/8/8/8/8/8/8/8 w - - 0 10";
 
-    return Position(fen, false).material_key();
+    return Position(fen, false, NULL).material_key();
   }
 
   template<typename M>

src/evaluate.cpp

@@ -371,7 +371,7 @@ Value do_evaluate(const Position& pos, Value& margin) {
   margins[WHITE] = margins[BLACK] = VALUE_ZERO;
 
   // Probe the material hash table
-  ei.mi = this_thread->materialTable.probe(pos);
+  ei.mi = pos.this_thread()->materialTable.probe(pos);
   score += ei.mi->material_value();
 
   // If we have a specialized evaluation function for the current material
@@ -383,7 +383,7 @@ Value do_evaluate(const Position& pos, Value& margin) {
   }
 
   // Probe the pawn hash table
-  ei.pi = this_thread->pawnTable.probe(pos);
+  ei.pi = pos.this_thread()->pawnTable.probe(pos);
   score += ei.pi->pawns_value();
 
   // Initialize attack and king safety bitboards

src/position.cpp

@@ -92,17 +92,19 @@ CheckInfo::CheckInfo(const Position& pos) {
 }
 
 
-/// Position::operator=() creates a copy of 'pos'. We want the new born Position
+/// Position::copy() creates a copy of 'pos'. We want the new born Position
 /// object do not depend on any external data so we detach state pointer from
 /// the source one.
 
-Position& Position::operator=(const Position& pos) {
+void Position::copy(const Position& pos, Thread* th) {
 
   memcpy(this, &pos, sizeof(Position));
   startState = *st;
   st = &startState;
+  thisThread = th;
   nodes = 0;
-  return *this;
+
+  assert(pos_is_ok());
 }
 
 
@@ -110,7 +112,7 @@ Position& Position::operator=(const Position& pos) {
 /// string. This function is not very robust - make sure that input FENs are
 /// correct (this is assumed to be the responsibility of the GUI).
 
-void Position::from_fen(const string& fenStr, bool isChess960) {
+void Position::from_fen(const string& fenStr, bool isChess960, Thread* th) {
 /*
    A FEN string defines a particular position using only the ASCII character set.
 
@@ -226,6 +228,7 @@ void Position::from_fen(const string& fenStr, bool isChess960) {
   st->npMaterial[BLACK] = compute_non_pawn_material(BLACK);
   st->checkersBB = attackers_to(king_square(sideToMove)) & pieces(~sideToMove);
   chess960 = isChess960;
+  thisThread = th;
 
   assert(pos_is_ok());
 }
@@ -328,7 +331,7 @@ void Position::print(Move move) const {
 
   if (move)
   {
-      Position p(*this);
+      Position p(*this, thisThread);
       cout << "\nMove is: " << (sideToMove == BLACK ? ".." : "") << move_to_san(p, move);
   }
 
@@ -895,8 +898,8 @@ void Position::do_move(Move m, StateInfo& newSt, const CheckInfo& ci, bool moveI
   }
 
   // Prefetch pawn and material hash tables
-  prefetch((char*)this_thread->pawnTable.entries[st->pawnKey]);
-  prefetch((char*)this_thread->materialTable.entries[st->materialKey]);
+  prefetch((char*)thisThread->pawnTable.entries[st->pawnKey]);
+  prefetch((char*)thisThread->materialTable.entries[st->materialKey]);
 
   // Update incremental scores
   st->psqScore += psq_delta(piece, from, to);
@@ -1538,9 +1541,10 @@ void Position::init() {
 void Position::flip() {
 
   // Make a copy of current position before to start changing
-  const Position pos(*this);
+  const Position pos(*this, thisThread);
 
   clear();
+  thisThread = pos.this_thread();
 
   // Board
   for (Square s = SQ_A1; s <= SQ_H8; s++)

src/position.h

@@ -84,14 +84,19 @@ struct StateInfo {
 ///    * A counter for detecting 50 move rule draws.
 
 class Position {
-public:
-  Position() {}
-  Position(const Position& p) { *this = p; }
-  Position(const std::string& f, bool c960) { from_fen(f, c960); }
+
+  // No copy c'tor or assignment operator allowed
+  Position(const Position&);
   Position& operator=(const Position&);
 
+public:
+  Position() {}
+  Position(const Position& p, Thread* t) { copy(p, t); }
+  Position(const std::string& f, bool c960, Thread* t) { from_fen(f, c960, t); }
+
   // Text input/output
-  void from_fen(const std::string& fen, bool isChess960);
+  void copy(const Position& pos, Thread* th);
+  void from_fen(const std::string& fen, bool isChess960, Thread* th);
   const std::string to_fen() const;
   void print(Move m = MOVE_NONE) const;
 
@@ -171,6 +176,7 @@ public:
   Color side_to_move() const;
   int startpos_ply_counter() const;
   bool is_chess960() const;
+  Thread* this_thread() const;
   int64_t nodes_searched() const;
   void set_nodes_searched(int64_t n);
   template<bool SkipRepetition> bool is_draw() const;
@@ -218,6 +224,7 @@ private:
   int64_t nodes;
   int startPosPly;
   Color sideToMove;
+  Thread* thisThread;
   StateInfo* st;
   int chess960;
 
@@ -427,4 +434,8 @@ inline PieceType Position::captured_piece_type() const {
   return st->capturedType;
 }
 
+inline Thread* Position::this_thread() const {
+  return thisThread;
+}
+
 #endif // !defined(POSITION_H_INCLUDED)

src/search.cpp

@@ -332,7 +332,7 @@ finalize:
   // but if we are pondering or in infinite search, we shouldn't print the best
   // move before we are told to do so.
   if (!Signals.stop && (Limits.ponder || Limits.infinite))
-      this_thread->wait_for_stop_or_ponderhit();
+      pos.this_thread()->wait_for_stop_or_ponderhit();
 
   // Best move could be MOVE_NONE when searching on a stalemate position
   cout << "bestmove " << move_to_uci(RootMoves[0].pv[0], Chess960)
@@ -543,7 +543,7 @@ namespace {
     bool isPvMove, inCheck, singularExtensionNode, givesCheck;
     bool captureOrPromotion, dangerous, doFullDepthSearch;
     int moveCount = 0, playedMoveCount = 0;
-    Thread* thisThread = this_thread;
+    Thread* thisThread = pos.this_thread();
     SplitPoint* sp = NULL;
 
     refinedValue = bestValue = value = -VALUE_INFINITE;
@@ -1825,7 +1825,7 @@ void Thread::idle_loop(SplitPoint* sp_master) {
           lock_release(Threads.splitLock);
 
           Stack ss[MAX_PLY_PLUS_2];
-          Position pos(*sp->pos);
+          Position pos(*sp->pos, this);
 
           memcpy(ss, sp->ss - 1, 4 * sizeof(Stack));
           (ss+1)->sp = sp;

src/thread.cpp

@@ -28,22 +28,17 @@
 using namespace Search;
 
 ThreadsManager Threads; // Global object
-THREAD_LOCAL Thread* this_thread; // Thread local variable
 
 namespace { extern "C" {
 
  // start_routine() is the C function which is called when a new thread
  // is launched. It is a wrapper to member function pointed by start_fn.
 
- long start_routine(Thread* th) {
-
-   this_thread = th; // Save pointer into thread local storage
-   (th->*(th->start_fn))();
-   return 0;
- }
+ long start_routine(Thread* th) { (th->*(th->start_fn))(); return 0; }
 
 } }
 
+
 // Thread c'tor starts a newly-created thread of execution that will call
 // the idle loop function pointed by start_fn going immediately to sleep.
 
@@ -210,7 +205,6 @@ void ThreadsManager::init() {
   lock_init(splitLock);
   timer = new Thread(&Thread::timer_loop);
   threads.push_back(new Thread(&Thread::main_loop));
-  this_thread = main_thread(); // Use main thread's resources
   read_uci_options();
 }
 
@@ -313,7 +307,7 @@ Value ThreadsManager::split(Position& pos, Stack* ss, Value alpha, Value beta,
   assert(beta <= VALUE_INFINITE);
   assert(depth > DEPTH_ZERO);
 
-  Thread* master = this_thread;
+  Thread* master = pos.this_thread();
 
   if (master->splitPointsCnt >= MAX_SPLITPOINTS_PER_THREAD)
       return bestValue;
@@ -440,7 +434,7 @@ void ThreadsManager::start_searching(const Position& pos, const LimitsType& limi
   Signals.stopOnPonderhit = Signals.firstRootMove = false;
   Signals.stop = Signals.failedLowAtRoot = false;
 
-  RootPosition = pos;
+  RootPosition.copy(pos, main_thread());
   Limits = limits;
   RootMoves.clear();
 

src/thread.h

@@ -119,7 +119,7 @@ public:
   bool use_sleeping_threads() const { return useSleepingThreads; }
   int min_split_depth() const { return minimumSplitDepth; }
   int size() const { return (int)threads.size(); }
-  Thread* main_thread() const { return threads[0]; }
+  Thread* main_thread() { return threads[0]; }
 
   void wake_up() const;
   void sleep() const;
@@ -146,6 +146,5 @@ private:
 };
 
 extern ThreadsManager Threads;
-extern THREAD_LOCAL Thread* this_thread;
 
 #endif // !defined(THREAD_H_INCLUDED)

src/types.h

@@ -64,12 +64,6 @@
 #  define FORCE_INLINE  inline
 #endif
 
-#if defined(__GNUC__)
-#  define THREAD_LOCAL  __thread
-#else
-#  define THREAD_LOCAL  __declspec(thread)
-#endif
-
 #if defined(USE_POPCNT)
 const bool HasPopCnt = true;
 #else

src/uci.cpp

@@ -56,7 +56,7 @@ namespace {
 
 void uci_loop(const string& args) {
 
-  Position pos(StartFEN, false); // The root position
+  Position pos(StartFEN, false, Threads.main_thread()); // The root position
   string cmd, token;
 
   while (token != "quit")
@@ -167,7 +167,7 @@ namespace {
     else
         return;
 
-    pos.from_fen(fen, Options["UCI_Chess960"]);
+    pos.from_fen(fen, Options["UCI_Chess960"], Threads.main_thread());
 
     // Parse move list (if any)
     while (is >> token && (m = move_from_uci(pos, token)) != MOVE_NONE)