/*
 
  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) 2015-2016 Marco Costalba, Joona Kiiski, Gary Linscott, Tord Romstad
 
 
 
  Stockfish is free software: you can redistribute it and/or modify
 
  it under the terms of the GNU General Public License as published by
 
  the Free Software Foundation, either version 3 of the License, or
 
  (at your option) any later version.
 
 
 
  Stockfish is distributed in the hope that it will be useful,
 
  but WITHOUT ANY WARRANTY; without even the implied warranty of
 
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 
  GNU General Public License for more details.
 
 
 
  You should have received a copy of the GNU General Public License
 
  along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
*/
 
 
 
#include <algorithm>
 
#include <cfloat>
 
#include <cmath>
 
 
 
#include "search.h"
 
#include "timeman.h"
 
#include "uci.h"
 
 
 
TimeManagement Time; // Our global time management object
 
 
 
namespace {
 
 
 
  enum TimeType { OptimumTime, MaxTime };
 
 
 
  const int MoveHorizon   = 50;   // Plan time management at most this many moves ahead
 
  const double MaxRatio   = 7.09;  // When in trouble, we can step over reserved time with this ratio
 
  const double StealRatio = 0.35; // However we must not steal time from remaining moves over this ratio
 
 
 
 
 
  // move_importance() is a skew-logistic function based on naive statistical
 
  // analysis of "how many games are still undecided after n half-moves". Game
 
  // is considered "undecided" as long as neither side has >275cp advantage.
 
  // Data was extracted from the CCRL game database with some simple filtering criteria.
 
 
 
  double move_importance(int ply) {
 
 
 
    const double XScale = 7.64;
 
    const double XShift = 58.4;
 
    const double Skew   = 0.183;
 
 
 
    return pow((1 + exp((ply - XShift) / XScale)), -Skew) + DBL_MIN; // Ensure non-zero
 
  }
 
 
 
  template<TimeType T>
 
  int remaining(int myTime, int movesToGo, int ply, int slowMover)
 
  {
 
    const double TMaxRatio   = (T == OptimumTime ? 1 : MaxRatio);
 
    const double TStealRatio = (T == OptimumTime ? 0 : StealRatio);
 
 
 
    double moveImportance = (move_importance(ply) * slowMover) / 100;
 
    double otherMovesImportance = 0;
 
 
 
    for (int i = 1; i < movesToGo; ++i)
 
        otherMovesImportance += move_importance(ply + 2 * i);
 
 
 
    double ratio1 = (TMaxRatio * moveImportance) / (TMaxRatio * moveImportance + otherMovesImportance);
 
    double ratio2 = (moveImportance + TStealRatio * otherMovesImportance) / (moveImportance + otherMovesImportance);
 
 
 
    return int(myTime * std::min(ratio1, ratio2)); // Intel C++ asks for an explicit cast
 
  }
 
 
 
} // namespace
 
 
 
 
 
/// init() is called at the beginning of the search and calculates the allowed
 
/// thinking time out of the time control and current game ply. We support four
 
/// different kinds of time controls, passed in 'limits':
 
///
 
///  inc == 0 && movestogo == 0 means: x basetime  [sudden death!]
 
///  inc == 0 && movestogo != 0 means: x moves in y minutes
 
///  inc >  0 && movestogo == 0 means: x basetime + z increment
 
///  inc >  0 && movestogo != 0 means: x moves in y minutes + z increment
 
 
 
void TimeManagement::init(Search::LimitsType& limits, Color us, int ply)
 
{
 
  int minThinkingTime = Options["Minimum Thinking Time"];
 
  int moveOverhead    = Options["Move Overhead"];
 
  int slowMover       = Options["Slow Mover"];
 
  int npmsec          = Options["nodestime"];
 
 
 
  // If we have to play in 'nodes as time' mode, then convert from time
 
  // to nodes, and use resulting values in time management formulas.
 
  // WARNING: Given npms (nodes per millisecond) must be much lower then
 
  // the real engine speed to avoid time losses.
 
  if (npmsec)
 
  {
 
      if (!availableNodes) // Only once at game start
 
          availableNodes = npmsec * limits.time[us]; // Time is in msec
 
 
 
      // Convert from millisecs to nodes
 
      limits.time[us] = (int)availableNodes;
 
      limits.inc[us] *= npmsec;
 
      limits.npmsec = npmsec;
 
  }
 
 
 
  startTime = limits.startTime;
 
  optimumTime = maximumTime = std::max(limits.time[us], minThinkingTime);
 
 
 
  const int MaxMTG = limits.movestogo ? std::min(limits.movestogo, MoveHorizon) : MoveHorizon;
 
 
 
  // We calculate optimum time usage for different hypothetical "moves to go"-values
 
  // and choose the minimum of calculated search time values. Usually the greatest
 
  // hypMTG gives the minimum values.
 
  for (int hypMTG = 1; hypMTG <= MaxMTG; ++hypMTG)
 
  {
 
      // Calculate thinking time for hypothetical "moves to go"-value
 
      int hypMyTime =  limits.time[us]
 
                     + limits.inc[us] * (hypMTG - 1)
 
                     - moveOverhead * (2 + std::min(hypMTG, 40));
 
 
 
      hypMyTime = std::max(hypMyTime, 0);
 
 
 
      int t1 = minThinkingTime + remaining<OptimumTime>(hypMyTime, hypMTG, ply, slowMover);
 
      int t2 = minThinkingTime + remaining<MaxTime    >(hypMyTime, hypMTG, ply, slowMover);
 
 
 
      optimumTime = std::min(t1, optimumTime);
 
      maximumTime = std::min(t2, maximumTime);
 
  }
 
 
 
  if (Options["Ponder"])
 
      optimumTime += optimumTime / 4;
 
}