Improve thread voting inefficiencies

Initialize the unordered map to a reasonable
number of buckets and make the move hashes well
distributed. For more see
#4958 (comment)
Also make bestThreadPV and newThreadPV references
so we don't copy entire vectors.

closes #5048

No functional change

src/thread.cpp

@@ -210,10 +210,9 @@ void ThreadPool::start_thinking(const OptionsMap&  options,
 
 Thread* ThreadPool::get_best_thread() const {
 
-    std::unordered_map<Move, int64_t, Move::MoveHash> votes;
-
     Thread* bestThread = threads.front();
     Value   minScore   = VALUE_NONE;
+    std::unordered_map<Move, int64_t, Move::MoveHash> votes(2 * std::min(size(), bestThread->worker->rootMoves.size())); 
 
     // Find the minimum score of all threads
     for (Thread* th : threads)
@@ -232,13 +231,12 @@ Thread* ThreadPool::get_best_thread() const {
         const auto bestThreadScore = bestThread->worker->rootMoves[0].score;
         const auto newThreadScore  = th->worker->rootMoves[0].score;
 
-        const auto bestThreadPV = bestThread->worker->rootMoves[0].pv;
-        const auto newThreadPV  = th->worker->rootMoves[0].pv;
+        const auto& bestThreadPV = bestThread->worker->rootMoves[0].pv;
+        const auto& newThreadPV  = th->worker->rootMoves[0].pv;
 
         const auto bestThreadMoveVote = votes[bestThreadPV[0]];
         const auto newThreadMoveVote  = votes[newThreadPV[0]];
 
-
         const bool bestThreadInProvenWin = bestThreadScore >= VALUE_TB_WIN_IN_MAX_PLY;
         const bool newThreadInProvenWin  = newThreadScore >= VALUE_TB_WIN_IN_MAX_PLY;
 

src/types.h

@@ -397,7 +397,7 @@ class Move {
     constexpr std::uint16_t raw() const { return data; }
 
     struct MoveHash {
-        std::size_t operator()(const Move& m) const { return m.data; }
+        std::size_t operator()(const Move& m) const { return make_key(m.data); }
     };
 
    protected: