synchronization.cc

Go to the documentation of this file.

// Copyright 2010-2018 Google LLC
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
 
#include "ortools/sat/synchronization.h"
 
#if !defined(__PORTABLE_PLATFORM__)
#include "ortools/base/file.h"
#include "ortools/sat/cp_model_loader.h"
#endif  // __PORTABLE_PLATFORM__
 
#include "absl/container/flat_hash_set.h"
#include "absl/random/random.h"
#include "ortools/base/integral_types.h"
#include "ortools/base/stl_util.h"
#include "ortools/sat/cp_model.pb.h"
#include "ortools/sat/cp_model_utils.h"
#include "ortools/sat/integer.h"
#include "ortools/sat/linear_programming_constraint.h"
#include "ortools/sat/model.h"
#include "ortools/sat/sat_base.h"
#include "ortools/util/time_limit.h"
 
ABSL_FLAG(bool, cp_model_dump_solutions, false,
          "DEBUG ONLY. If true, all the intermediate solution will be dumped "
          "under '\"FLAGS_cp_model_dump_prefix\" + \"solution_xxx.pb.txt\"'.");
 
ABSL_FLAG(
    std::string, cp_model_load_debug_solution, "",
    "DEBUG ONLY. When this is set to a non-empty file name, "
    "we will interpret this as an internal solution which can be used for "
    "debugging. For instance we use it to identify wrong cuts/reasons.");
 
namespace operations_research {
namespace sat {
 
void SharedRelaxationSolutionRepository::NewRelaxationSolution(
    const CpSolverResponse& response) {
  // Note that the Add() method already applies mutex lock. So we don't need it
  // here.
  if (response.solution().empty()) return;
 
  // Add this solution to the pool.
  SharedSolutionRepository<int64>::Solution solution;
  solution.variable_values.assign(response.solution().begin(),
                                  response.solution().end());
  // For now we use the negated lower bound as the "internal objective" to
  // prefer solution with an higher bound.
  //
  // Note: If the model doesn't have objective, the best_objective_bound is set
  // to default value 0.
  solution.rank = -response.best_objective_bound();
 
  Add(solution);
}
 
void SharedLPSolutionRepository::NewLPSolution(
    std::vector<double> lp_solution) {
  if (lp_solution.empty()) return;
 
  // Add this solution to the pool.
  SharedSolutionRepository<double>::Solution solution;
  solution.variable_values = std::move(lp_solution);
 
  // We always prefer to keep the solution from the last synchronize batch.
  absl::MutexLock mutex_lock(&mutex_);
  solution.rank = -num_synchronization_;
  AddInternal(solution);
}
 
bool SharedIncompleteSolutionManager::HasNewSolution() const {
  absl::MutexLock mutex_lock(&mutex_);
  return !solutions_.empty();
}
 
std::vector<double> SharedIncompleteSolutionManager::GetNewSolution() {
  absl::MutexLock mutex_lock(&mutex_);
  std::vector<double> solution;
  if (solutions_.empty()) return solution;
 
  solution = std::move(solutions_.back());
  solutions_.pop_back();
  return solution;
}
 
void SharedIncompleteSolutionManager::AddNewSolution(
    const std::vector<double>& lp_solution) {
  absl::MutexLock mutex_lock(&mutex_);
  solutions_.push_back(lp_solution);
}
 
// TODO(user): Experiments and play with the num_solutions_to_keep parameter.
SharedResponseManager::SharedResponseManager(bool log_updates,
                                             bool enumerate_all_solutions,
                                             const CpModelProto* proto,
                                             const WallTimer* wall_timer,
                                             SharedTimeLimit* shared_time_limit)
    : log_updates_(log_updates),
      enumerate_all_solutions_(enumerate_all_solutions),
      model_proto_(*proto),
      wall_timer_(*wall_timer),
      shared_time_limit_(shared_time_limit),
      solutions_(/*num_solutions_to_keep=*/3) {}
 
namespace {
 
void LogNewSolution(const std::string& event_or_solution_count,
                    double time_in_seconds, double obj_best, double obj_lb,
                    double obj_ub, const std::string& solution_info) {
  const std::string obj_next =
      absl::StrFormat("next:[%.9g,%.9g]", obj_lb, obj_ub);
  LOG(INFO) << absl::StrFormat("#%-5s %6.2fs best:%-5.9g %-15s %s",
                               event_or_solution_count, time_in_seconds,
                               obj_best, obj_next, solution_info);
}
 
void LogNewSatSolution(const std::string& event_or_solution_count,
                       double time_in_seconds,
                       const std::string& solution_info) {
  LOG(INFO) << absl::StrFormat("#%-5s %6.2fs  %s", event_or_solution_count,
                               time_in_seconds, solution_info);
}
 
}  // namespace
 
void SharedResponseManager::UpdatePrimalIntegral() {
  absl::MutexLock mutex_lock(&mutex_);
  if (!model_proto_.has_objective()) return;
 
  const double current_time = shared_time_limit_->GetElapsedDeterministicTime();
  const double time_delta = current_time - last_primal_integral_time_stamp_;
  last_primal_integral_time_stamp_ = current_time;
 
  // We use the log of the absolute objective gap.
  //
  // Using the log should count no solution as just log(2*64) = 18, and
  // otherwise just compare order of magnitude which seems nice. Also, It is
  // more easy to compare the primal integral with the total time.
  const CpObjectiveProto& obj = model_proto_.objective();
  const double factor =
      obj.scaling_factor() != 0.0 ? std::abs(obj.scaling_factor()) : 1.0;
  const double bounds_delta = std::log(
      1 + factor * std::abs(static_cast<double>(inner_objective_upper_bound_) -
                            static_cast<double>(inner_objective_lower_bound_)));
  primal_integral_ += time_delta * bounds_delta;
}
 
void SharedResponseManager::SetGapLimitsFromParameters(
    const SatParameters& parameters) {
  absl::MutexLock mutex_lock(&mutex_);
  if (!model_proto_.has_objective()) return;
  absolute_gap_limit_ = parameters.absolute_gap_limit();
  relative_gap_limit_ = parameters.relative_gap_limit();
}
 
void SharedResponseManager::TestGapLimitsIfNeeded() {
  if (absolute_gap_limit_ == 0 && relative_gap_limit_ == 0) return;
  if (best_solution_objective_value_ >= kMaxIntegerValue) return;
  if (inner_objective_lower_bound_ <= kMinIntegerValue) return;
 
  const CpObjectiveProto& obj = model_proto_.objective();
  const double user_best =
      ScaleObjectiveValue(obj, best_solution_objective_value_);
  const double user_bound =
      ScaleObjectiveValue(obj, inner_objective_lower_bound_);
  const double gap = std::abs(user_best - user_bound);
  if (gap <= absolute_gap_limit_) {
    LOG_IF(INFO, log_updates_)
        << "Absolute gap limit of " << absolute_gap_limit_ << " reached.";
    best_response_.set_status(CpSolverStatus::OPTIMAL);
 
    // Note(user): Some code path in single-thread assumes that the problem
    // can only be solved when they have proven infeasibility and do not check
    // the ProblemIsSolved() method. So we force a stop here.
    shared_time_limit_->Stop();
  }
  if (gap / std::max(1.0, std::abs(user_best)) < relative_gap_limit_) {
    LOG_IF(INFO, log_updates_)
        << "Relative gap limit of " << relative_gap_limit_ << " reached.";
    best_response_.set_status(CpSolverStatus::OPTIMAL);
 
    // Same as above.
    shared_time_limit_->Stop();
  }
}
 
void SharedResponseManager::UpdateInnerObjectiveBounds(
    const std::string& worker_info, IntegerValue lb, IntegerValue ub) {
  absl::MutexLock mutex_lock(&mutex_);
  CHECK(model_proto_.has_objective());
 
  // The problem is already solved!
  //
  // TODO(user): A thread might not be notified right away that the new bounds
  // that it is pushing make the problem infeasible. Fix that. For now we just
  // abort early here to avoid logging the "#Done" message multiple times.
  if (inner_objective_lower_bound_ > inner_objective_upper_bound_) {
    return;
  }
 
  const bool change =
      (lb > inner_objective_lower_bound_ || ub < inner_objective_upper_bound_);
  if (lb > inner_objective_lower_bound_) {
    // When the improving problem is infeasible, it is possible to report
    // arbitrary high inner_objective_lower_bound_. We make sure it never cross
    // the current best solution, so that we always report globablly valid lower
    // bound.
    DCHECK_LE(inner_objective_upper_bound_, best_solution_objective_value_);
    inner_objective_lower_bound_ =
        std::min(best_solution_objective_value_, lb.value());
  }
  if (ub < inner_objective_upper_bound_) {
    inner_objective_upper_bound_ = ub.value();
  }
  if (inner_objective_lower_bound_ > inner_objective_upper_bound_) {
    if (best_response_.status() == CpSolverStatus::FEASIBLE ||
        best_response_.status() == CpSolverStatus::OPTIMAL) {
      best_response_.set_status(CpSolverStatus::OPTIMAL);
    } else {
      best_response_.set_status(CpSolverStatus::INFEASIBLE);
    }
    if (log_updates_) LogNewSatSolution("Done", wall_timer_.Get(), worker_info);
    return;
  }
  if (log_updates_ && change) {
    const CpObjectiveProto& obj = model_proto_.objective();
    const double best =
        ScaleObjectiveValue(obj, best_solution_objective_value_);
    double new_lb = ScaleObjectiveValue(obj, inner_objective_lower_bound_);
    double new_ub = ScaleObjectiveValue(obj, inner_objective_upper_bound_);
    if (model_proto_.objective().scaling_factor() < 0) {
      std::swap(new_lb, new_ub);
    }
    LogNewSolution("Bound", wall_timer_.Get(), best, new_lb, new_ub,
                   worker_info);
  }
  if (change) TestGapLimitsIfNeeded();
}
 
// Invariant: the status always start at UNKNOWN and can only evolve as follow:
// UNKNOWN -> FEASIBLE -> OPTIMAL
// UNKNOWN -> INFEASIBLE
void SharedResponseManager::NotifyThatImprovingProblemIsInfeasible(
    const std::string& worker_info) {
  absl::MutexLock mutex_lock(&mutex_);
  if (best_response_.status() == CpSolverStatus::FEASIBLE ||
      best_response_.status() == CpSolverStatus::OPTIMAL) {
    // We also use this status to indicate that we enumerated all solutions to
    // a feasible problem.
    best_response_.set_status(CpSolverStatus::OPTIMAL);
    if (!model_proto_.has_objective()) {
      best_response_.set_all_solutions_were_found(true);
    }
 
    // We just proved that the best solution cannot be improved uppon, so we
    // have a new lower bound.
    inner_objective_lower_bound_ = best_solution_objective_value_;
  } else {
    CHECK_EQ(num_solutions_, 0);
    best_response_.set_status(CpSolverStatus::INFEASIBLE);
  }
  if (log_updates_) LogNewSatSolution("Done", wall_timer_.Get(), worker_info);
}
 
void SharedResponseManager::AddUnsatCore(const std::vector<int>& core) {
  absl::MutexLock mutex_lock(&mutex_);
  best_response_.clear_sufficient_assumptions_for_infeasibility();
  for (const int ref : core) {
    best_response_.add_sufficient_assumptions_for_infeasibility(ref);
  }
}
 
IntegerValue SharedResponseManager::GetInnerObjectiveLowerBound() {
  absl::MutexLock mutex_lock(&mutex_);
  return IntegerValue(inner_objective_lower_bound_);
}
 
IntegerValue SharedResponseManager::GetInnerObjectiveUpperBound() {
  absl::MutexLock mutex_lock(&mutex_);
  return IntegerValue(inner_objective_upper_bound_);
}
 
void SharedResponseManager::Synchronize() {
  absl::MutexLock mutex_lock(&mutex_);
  synchronized_inner_objective_lower_bound_ =
      IntegerValue(inner_objective_lower_bound_);
  synchronized_inner_objective_upper_bound_ =
      IntegerValue(inner_objective_upper_bound_);
}
 
IntegerValue SharedResponseManager::SynchronizedInnerObjectiveLowerBound() {
  absl::MutexLock mutex_lock(&mutex_);
  return synchronized_inner_objective_lower_bound_;
}
 
IntegerValue SharedResponseManager::SynchronizedInnerObjectiveUpperBound() {
  absl::MutexLock mutex_lock(&mutex_);
  return synchronized_inner_objective_upper_bound_;
}
 
IntegerValue SharedResponseManager::BestSolutionInnerObjectiveValue() {
  absl::MutexLock mutex_lock(&mutex_);
  return IntegerValue(best_solution_objective_value_);
}
 
double SharedResponseManager::PrimalIntegral() const {
  absl::MutexLock mutex_lock(&mutex_);
  return primal_integral_;
}
 
int SharedResponseManager::AddSolutionCallback(
    std::function<void(const CpSolverResponse&)> callback) {
  absl::MutexLock mutex_lock(&mutex_);
  const int id = next_callback_id_++;
  callbacks_.emplace_back(id, std::move(callback));
  return id;
}
 
void SharedResponseManager::UnregisterCallback(int callback_id) {
  absl::MutexLock mutex_lock(&mutex_);
  for (int i = 0; i < callbacks_.size(); ++i) {
    if (callbacks_[i].first == callback_id) {
      callbacks_.erase(callbacks_.begin() + i);
      return;
    }
  }
  LOG(DFATAL) << "Callback id " << callback_id << " not registered.";
}
 
CpSolverResponse SharedResponseManager::GetResponse() {
  absl::MutexLock mutex_lock(&mutex_);
  FillObjectiveValuesInBestResponse();
  return best_response_;
}
 
void SharedResponseManager::FillObjectiveValuesInBestResponse() {
  if (!model_proto_.has_objective()) return;
  const CpObjectiveProto& obj = model_proto_.objective();
 
  if (best_response_.status() == CpSolverStatus::INFEASIBLE) {
    best_response_.clear_objective_value();
    best_response_.clear_best_objective_bound();
    return;
  }
 
  // Set the objective value.
  // If we don't have any solution, we use our inner bound.
  if (best_response_.status() == CpSolverStatus::UNKNOWN) {
    best_response_.set_objective_value(
        ScaleObjectiveValue(obj, inner_objective_upper_bound_));
  } else {
    best_response_.set_objective_value(
        ScaleObjectiveValue(obj, best_solution_objective_value_));
  }
 
  // Update the best bound in the response.
  best_response_.set_best_objective_bound(
      ScaleObjectiveValue(obj, inner_objective_lower_bound_));
 
  // Update the primal integral.
  best_response_.set_primal_integral(primal_integral_);
}
 
void SharedResponseManager::NewSolution(const CpSolverResponse& response,
                                        Model* model) {
  absl::MutexLock mutex_lock(&mutex_);
 
  if (model_proto_.has_objective()) {
    const int64 objective_value =
        ComputeInnerObjective(model_proto_.objective(), response);
 
    // Add this solution to the pool, even if it is not improving.
    if (!response.solution().empty()) {
      SharedSolutionRepository<int64>::Solution solution;
      solution.variable_values.assign(response.solution().begin(),
                                      response.solution().end());
      solution.rank = objective_value;
      solutions_.Add(solution);
    }
 
    // Ignore any non-strictly improving solution.
    if (objective_value > inner_objective_upper_bound_) return;
 
    // Our inner_objective_lower_bound_ should be a globaly valid bound, until
    // the problem become infeasible (i.e the lb > ub) in which case the bound
    // is no longer globally valid. Here, because we have a strictly improving
    // solution, we shouldn't be in the infeasible setting yet.
    DCHECK_GE(objective_value, inner_objective_lower_bound_);
 
    DCHECK_LT(objective_value, best_solution_objective_value_);
    best_solution_objective_value_ = objective_value;
 
    // Update the new bound.
    inner_objective_upper_bound_ = objective_value - 1;
  }
 
  // Note that the objective will be filled by
  // FillObjectiveValuesInBestResponse().
  if (!model_proto_.has_objective() && !enumerate_all_solutions_) {
    best_response_.set_status(CpSolverStatus::OPTIMAL);
  } else {
    best_response_.set_status(CpSolverStatus::FEASIBLE);
  }
 
  best_response_.set_solution_info(response.solution_info());
  *best_response_.mutable_solution() = response.solution();
  *best_response_.mutable_solution_lower_bounds() =
      response.solution_lower_bounds();
  *best_response_.mutable_solution_upper_bounds() =
      response.solution_upper_bounds();
 
  // Mark model as OPTIMAL if the inner bound crossed.
  if (model_proto_.has_objective() &&
      inner_objective_lower_bound_ > inner_objective_upper_bound_) {
    best_response_.set_status(CpSolverStatus::OPTIMAL);
  }
 
  // Logging.
  ++num_solutions_;
  if (log_updates_) {
    std::string solution_info = response.solution_info();
    if (model != nullptr) {
      const int64 num_bool = model->Get<Trail>()->NumVariables();
      const int64 num_fixed = model->Get<SatSolver>()->NumFixedVariables();
      absl::StrAppend(&solution_info, " fixed_bools:", num_fixed, "/",
                      num_bool);
    }
 
    if (model_proto_.has_objective()) {
      const CpObjectiveProto& obj = model_proto_.objective();
      const double best =
          ScaleObjectiveValue(obj, best_solution_objective_value_);
      double lb = ScaleObjectiveValue(obj, inner_objective_lower_bound_);
      double ub = ScaleObjectiveValue(obj, inner_objective_upper_bound_);
      if (model_proto_.objective().scaling_factor() < 0) {
        std::swap(lb, ub);
      }
      LogNewSolution(absl::StrCat(num_solutions_), wall_timer_.Get(), best, lb,
                     ub, solution_info);
    } else {
      LogNewSatSolution(absl::StrCat(num_solutions_), wall_timer_.Get(),
                        solution_info);
    }
  }
 
  // Call callbacks.
  // Note that we cannot call function that try to get the mutex_ here.
  TestGapLimitsIfNeeded();
  if (!callbacks_.empty()) {
    FillObjectiveValuesInBestResponse();
    SetStatsFromModelInternal(model);
    for (const auto& pair : callbacks_) {
      pair.second(best_response_);
    }
  }
 
#if !defined(__PORTABLE_PLATFORM__)
  // We protect solution dumping with log_updates as LNS subsolvers share
  // another solution manager, and we do not want to dump those.
  if (absl::GetFlag(FLAGS_cp_model_dump_solutions) && log_updates_) {
    const std::string file =
        absl::StrCat(dump_prefix_, "solution_", num_solutions_, ".pbtxt");
    LOG(INFO) << "Dumping solution to '" << file << "'.";
    CHECK_OK(file::SetTextProto(file, best_response_, file::Defaults()));
  }
#endif  // __PORTABLE_PLATFORM__
}
 
void SharedResponseManager::LoadDebugSolution(Model* model) {
#if !defined(__PORTABLE_PLATFORM__)
  if (absl::GetFlag(FLAGS_cp_model_load_debug_solution).empty()) return;
  if (model->Get<DebugSolution>() != nullptr) return;  // Already loaded.
 
  CpSolverResponse response;
  LOG(INFO) << "Reading solution from '"
            << absl::GetFlag(FLAGS_cp_model_load_debug_solution) << "'.";
  CHECK_OK(file::GetTextProto(absl::GetFlag(FLAGS_cp_model_load_debug_solution),
                              &response, file::Defaults()));
 
  const auto& mapping = *model->GetOrCreate<CpModelMapping>();
  auto& debug_solution = *model->GetOrCreate<DebugSolution>();
  debug_solution.resize(
      model->GetOrCreate<IntegerTrail>()->NumIntegerVariables().value());
  for (int i = 0; i < response.solution().size(); ++i) {
    if (!mapping.IsInteger(i)) continue;
    const IntegerVariable var = mapping.Integer(i);
    debug_solution[var] = response.solution(i);
    debug_solution[NegationOf(var)] = -response.solution(i);
  }
 
  // The objective variable is usually not part of the proto, but it is still
  // nice to have it, so we recompute it here.
  auto* objective_def = model->Get<ObjectiveDefinition>();
  if (objective_def == nullptr) return;
 
  const IntegerVariable objective_var = objective_def->objective_var;
  const int64 objective_value =
      ComputeInnerObjective(model_proto_.objective(), response);
  debug_solution[objective_var] = objective_value;
  debug_solution[NegationOf(objective_var)] = -objective_value;
#endif  // __PORTABLE_PLATFORM__
}
 
void SharedResponseManager::SetStatsFromModel(Model* model) {
  absl::MutexLock mutex_lock(&mutex_);
  SetStatsFromModelInternal(model);
}
 
void SharedResponseManager::SetStatsFromModelInternal(Model* model) {
  if (model == nullptr) return;
  auto* sat_solver = model->Get<SatSolver>();
  auto* integer_trail = model->Get<IntegerTrail>();
  best_response_.set_num_booleans(sat_solver->NumVariables());
  best_response_.set_num_branches(sat_solver->num_branches());
  best_response_.set_num_conflicts(sat_solver->num_failures());
  best_response_.set_num_binary_propagations(sat_solver->num_propagations());
  best_response_.set_num_restarts(sat_solver->num_restarts());
  best_response_.set_num_integer_propagations(
      integer_trail == nullptr ? 0 : integer_trail->num_enqueues());
  auto* time_limit = model->Get<TimeLimit>();
  best_response_.set_wall_time(time_limit->GetElapsedTime());
  best_response_.set_deterministic_time(
      time_limit->GetElapsedDeterministicTime());
 
  int64 num_lp_iters = 0;
  for (const LinearProgrammingConstraint* lp :
       *model->GetOrCreate<LinearProgrammingConstraintCollection>()) {
    num_lp_iters += lp->total_num_simplex_iterations();
  }
  best_response_.set_num_lp_iterations(num_lp_iters);
}
 
bool SharedResponseManager::ProblemIsSolved() const {
  absl::MutexLock mutex_lock(&mutex_);
  return best_response_.status() == CpSolverStatus::OPTIMAL ||
         best_response_.status() == CpSolverStatus::INFEASIBLE;
}
 
SharedBoundsManager::SharedBoundsManager(const CpModelProto& model_proto)
    : num_variables_(model_proto.variables_size()),
      model_proto_(model_proto),
      lower_bounds_(num_variables_, kint64min),
      upper_bounds_(num_variables_, kint64max),
      synchronized_lower_bounds_(num_variables_, kint64min),
      synchronized_upper_bounds_(num_variables_, kint64max) {
  changed_variables_since_last_synchronize_.ClearAndResize(num_variables_);
  for (int i = 0; i < num_variables_; ++i) {
    lower_bounds_[i] = model_proto.variables(i).domain(0);
    const int domain_size = model_proto.variables(i).domain_size();
    upper_bounds_[i] = model_proto.variables(i).domain(domain_size - 1);
    synchronized_lower_bounds_[i] = lower_bounds_[i];
    synchronized_upper_bounds_[i] = upper_bounds_[i];
  }
}
 
void SharedBoundsManager::ReportPotentialNewBounds(
    const CpModelProto& model_proto, const std::string& worker_name,
    const std::vector<int>& variables,
    const std::vector<int64>& new_lower_bounds,
    const std::vector<int64>& new_upper_bounds) {
  CHECK_EQ(variables.size(), new_lower_bounds.size());
  CHECK_EQ(variables.size(), new_upper_bounds.size());
  int num_improvements = 0;
 
  absl::MutexLock mutex_lock(&mutex_);
  for (int i = 0; i < variables.size(); ++i) {
    const int var = variables[i];
    if (var >= num_variables_) continue;
    const int64 old_lb = lower_bounds_[var];
    const int64 old_ub = upper_bounds_[var];
    const int64 new_lb = new_lower_bounds[i];
    const int64 new_ub = new_upper_bounds[i];
    const bool changed_lb = new_lb > old_lb;
    const bool changed_ub = new_ub < old_ub;
    CHECK_GE(var, 0);
    if (!changed_lb && !changed_ub) continue;
 
    if (changed_lb) {
      lower_bounds_[var] = new_lb;
    }
    if (changed_ub) {
      upper_bounds_[var] = new_ub;
    }
    changed_variables_since_last_synchronize_.Set(var);
    num_improvements++;
  }
  // TODO(user): Display number of bound improvements cumulatively per
  // workers at the end of the search.
  if (num_improvements > 0) {
    VLOG(2) << worker_name << " exports " << num_improvements
            << " modifications";
  }
}
 
void SharedBoundsManager::Synchronize() {
  absl::MutexLock mutex_lock(&mutex_);
  for (const int var :
       changed_variables_since_last_synchronize_.PositionsSetAtLeastOnce()) {
    synchronized_lower_bounds_[var] = lower_bounds_[var];
    synchronized_upper_bounds_[var] = upper_bounds_[var];
    for (int j = 0; j < id_to_changed_variables_.size(); ++j) {
      id_to_changed_variables_[j].Set(var);
    }
  }
  changed_variables_since_last_synchronize_.ClearAll();
}
 
int SharedBoundsManager::RegisterNewId() {
  absl::MutexLock mutex_lock(&mutex_);
  const int id = id_to_changed_variables_.size();
  id_to_changed_variables_.resize(id + 1);
  id_to_changed_variables_[id].ClearAndResize(num_variables_);
  for (int var = 0; var < num_variables_; ++var) {
    const int64 lb = model_proto_.variables(var).domain(0);
    const int domain_size = model_proto_.variables(var).domain_size();
    const int64 ub = model_proto_.variables(var).domain(domain_size - 1);
    if (lb != synchronized_lower_bounds_[var] ||
        ub != synchronized_upper_bounds_[var]) {
      id_to_changed_variables_[id].Set(var);
    }
  }
  return id;
}
 
void SharedBoundsManager::GetChangedBounds(
    int id, std::vector<int>* variables, std::vector<int64>* new_lower_bounds,
    std::vector<int64>* new_upper_bounds) {
  variables->clear();
  new_lower_bounds->clear();
  new_upper_bounds->clear();
 
  absl::MutexLock mutex_lock(&mutex_);
  for (const int var : id_to_changed_variables_[id].PositionsSetAtLeastOnce()) {
    variables->push_back(var);
    new_lower_bounds->push_back(synchronized_lower_bounds_[var]);
    new_upper_bounds->push_back(synchronized_upper_bounds_[var]);
  }
  id_to_changed_variables_[id].ClearAll();
}
 
}  // namespace sat
}  // namespace operations_research