all_different.cc

Go to the documentation of this file.

// Copyright 2010-2021 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/all_different.h"

#include <algorithm>
#include <cstdint>
#include <limits>
#include <map>
#include <memory>

#include "absl/container/flat_hash_set.h"
#include "ortools/base/int_type.h"
#include "ortools/base/logging.h"
#include "ortools/base/map_util.h"
#include "ortools/graph/strongly_connected_components.h"
#include "ortools/sat/sat_solver.h"
#include "ortools/util/sort.h"

namespace operations_research {
namespace sat {

std::function<void(Model*)> AllDifferentBinary(
    const std::vector<IntegerVariable>& vars) {
  return [=](Model* model) {
    // Fully encode all the given variables and construct a mapping value ->
    // List of literal each indicating that a given variable takes this value.
    //
    // Note that we use a map to always add the constraints in the same order.
    std::map<IntegerValue, std::vector<Literal>> value_to_literals;
    IntegerEncoder* encoder = model->GetOrCreate<IntegerEncoder>();
    for (const IntegerVariable var : vars) {
      model->Add(FullyEncodeVariable(var));
      for (const auto& entry : encoder->FullDomainEncoding(var)) {
        value_to_literals[entry.value].push_back(entry.literal);
      }
    }

    // Add an at most one constraint for each value.
    for (const auto& entry : value_to_literals) {
      if (entry.second.size() > 1) {
        model->Add(AtMostOneConstraint(entry.second));
      }
    }

    // If the number of values is equal to the number of variables, we have
    // a permutation. We can add a bool_or for each literals attached to a
    // value.
    if (value_to_literals.size() == vars.size()) {
      for (const auto& entry : value_to_literals) {
        model->Add(ClauseConstraint(entry.second));
      }
    }
  };
}

std::function<void(Model*)> AllDifferentOnBounds(
    const std::vector<AffineExpression>& expressions) {
  return [=](Model* model) {
    if (expressions.empty()) return;
    auto* constraint = new AllDifferentBoundsPropagator(
        expressions, model->GetOrCreate<IntegerTrail>());
    constraint->RegisterWith(model->GetOrCreate<GenericLiteralWatcher>());
    model->TakeOwnership(constraint);
  };
}

std::function<void(Model*)> AllDifferentOnBounds(
    const std::vector<IntegerVariable>& vars) {
  return [=](Model* model) {
    if (vars.empty()) return;
    std::vector<AffineExpression> expressions;
    expressions.reserve(vars.size());
    for (const IntegerVariable var : vars) {
      expressions.push_back(AffineExpression(var));
    }
    auto* constraint = new AllDifferentBoundsPropagator(
        expressions, model->GetOrCreate<IntegerTrail>());
    constraint->RegisterWith(model->GetOrCreate<GenericLiteralWatcher>());
    model->TakeOwnership(constraint);
  };
}

std::function<void(Model*)> AllDifferentAC(
    const std::vector<IntegerVariable>& variables) {
  return [=](Model* model) {
    if (variables.size() < 3) return;

    AllDifferentConstraint* constraint = new AllDifferentConstraint(
        variables, model->GetOrCreate<IntegerEncoder>(),
        model->GetOrCreate<Trail>(), model->GetOrCreate<IntegerTrail>());
    constraint->RegisterWith(model->GetOrCreate<GenericLiteralWatcher>());
    model->TakeOwnership(constraint);
  };
}

AllDifferentConstraint::AllDifferentConstraint(
    std::vector<IntegerVariable> variables, IntegerEncoder* encoder,
    Trail* trail, IntegerTrail* integer_trail)
    : num_variables_(variables.size()),
      variables_(std::move(variables)),
      trail_(trail),
      integer_trail_(integer_trail) {
  // Initialize literals cache.
  int64_t min_value = std::numeric_limits<int64_t>::max();
  int64_t max_value = std::numeric_limits<int64_t>::min();
  variable_min_value_.resize(num_variables_);
  variable_max_value_.resize(num_variables_);
  variable_literal_index_.resize(num_variables_);
  int num_fixed_variables = 0;
  for (int x = 0; x < num_variables_; x++) {
    variable_min_value_[x] = integer_trail_->LowerBound(variables_[x]).value();
    variable_max_value_[x] = integer_trail_->UpperBound(variables_[x]).value();

    // Compute value range of all variables.
    min_value = std::min(min_value, variable_min_value_[x]);
    max_value = std::max(max_value, variable_max_value_[x]);

    // FullyEncode does not like 1-value domains, handle this case first.
    // TODO(user): Prune now, ignore these variables during solving.
    if (variable_min_value_[x] == variable_max_value_[x]) {
      num_fixed_variables++;
      variable_literal_index_[x].push_back(kTrueLiteralIndex);
      continue;
    }

    // Force full encoding if not already done.
    if (!encoder->VariableIsFullyEncoded(variables_[x])) {
      encoder->FullyEncodeVariable(variables_[x]);
    }

    // Fill cache with literals, default value is kFalseLiteralIndex.
    int64_t size = variable_max_value_[x] - variable_min_value_[x] + 1;
    variable_literal_index_[x].resize(size, kFalseLiteralIndex);
    for (const auto& entry : encoder->FullDomainEncoding(variables_[x])) {
      int64_t value = entry.value.value();
      // Can happen because of initial propagation!
      if (value < variable_min_value_[x] || variable_max_value_[x] < value) {
        continue;
      }
      variable_literal_index_[x][value - variable_min_value_[x]] =
          entry.literal.Index();
    }
  }
  min_all_values_ = min_value;
  num_all_values_ = max_value - min_value + 1;

  successor_.resize(num_variables_);
  variable_to_value_.assign(num_variables_, -1);
  visiting_.resize(num_variables_);
  variable_visited_from_.resize(num_variables_);
  residual_graph_successors_.resize(num_variables_ + num_all_values_ + 1);
  component_number_.resize(num_variables_ + num_all_values_ + 1);
}

void AllDifferentConstraint::RegisterWith(GenericLiteralWatcher* watcher) {
  const int id = watcher->Register(this);
  watcher->SetPropagatorPriority(id, 2);
  for (const auto& literal_indices : variable_literal_index_) {
    for (const LiteralIndex li : literal_indices) {
      // Watch only unbound literals.
      if (li >= 0 &&
          !trail_->Assignment().VariableIsAssigned(Literal(li).Variable())) {
        watcher->WatchLiteral(Literal(li), id);
        watcher->WatchLiteral(Literal(li).Negated(), id);
      }
    }
  }
}

LiteralIndex AllDifferentConstraint::VariableLiteralIndexOf(int x,
                                                            int64_t value) {
  return (value < variable_min_value_[x] || variable_max_value_[x] < value)
             ? kFalseLiteralIndex
             : variable_literal_index_[x][value - variable_min_value_[x]];
}

inline bool AllDifferentConstraint::VariableHasPossibleValue(int x,
                                                             int64_t value) {
  LiteralIndex li = VariableLiteralIndexOf(x, value);
  if (li == kFalseLiteralIndex) return false;
  if (li == kTrueLiteralIndex) return true;
  DCHECK_GE(li, 0);
  return !trail_->Assignment().LiteralIsFalse(Literal(li));
}

bool AllDifferentConstraint::MakeAugmentingPath(int start) {
  // Do a BFS and use visiting_ as a queue, with num_visited pointing
  // at its begin() and num_to_visit its end().
  // To switch to the augmenting path once a nonmatched value was found,
  // we remember the BFS tree in variable_visited_from_.
  int num_to_visit = 0;
  int num_visited = 0;
  // Enqueue start.
  visiting_[num_to_visit++] = start;
  variable_visited_[start] = true;
  variable_visited_from_[start] = -1;

  while (num_visited < num_to_visit) {
    // Dequeue node to visit.
    const int node = visiting_[num_visited++];

    for (const int value : successor_[node]) {
      if (value_visited_[value]) continue;
      value_visited_[value] = true;
      if (value_to_variable_[value] == -1) {
        // value is not matched: change path from node to start, and return.
        int path_node = node;
        int path_value = value;
        while (path_node != -1) {
          int old_value = variable_to_value_[path_node];
          variable_to_value_[path_node] = path_value;
          value_to_variable_[path_value] = path_node;
          path_node = variable_visited_from_[path_node];
          path_value = old_value;
        }
        return true;
      } else {
        // Enqueue node matched to value.
        const int next_node = value_to_variable_[value];
        variable_visited_[next_node] = true;
        visiting_[num_to_visit++] = next_node;
        variable_visited_from_[next_node] = node;
      }
    }
  }
  return false;
}

// The algorithm copies the solver state to successor_, which is used to compute
// a matching. If all variables can be matched, it generates the residual graph
// in separate vectors, computes its SCCs, and filters variable -> value if
// variable is not in the same SCC as value.
// Explanations for failure and filtering are fine-grained:
// failure is explained by a Hall set, i.e. dom(variables) \subseteq {values},
// with |variables| < |values|; filtering is explained by the Hall set that
// would happen if the variable was assigned to the value.
//
// TODO(user): If needed, there are several ways performance could be
// improved.
// If copying the variable state is too costly, it could be maintained instead.
// If the propagator has too many fruitless calls (without failing/pruning),
// we can remember the O(n) arcs used in the matching and the SCC decomposition,
// and guard calls to Propagate() if these arcs are still valid.
bool AllDifferentConstraint::Propagate() {
  // Copy variable state to graph state.
  prev_matching_ = variable_to_value_;
  value_to_variable_.assign(num_all_values_, -1);
  variable_to_value_.assign(num_variables_, -1);
  for (int x = 0; x < num_variables_; x++) {
    successor_[x].clear();
    const int64_t min_value = integer_trail_->LowerBound(variables_[x]).value();
    const int64_t max_value = integer_trail_->UpperBound(variables_[x]).value();
    for (int64_t value = min_value; value <= max_value; value++) {
      if (VariableHasPossibleValue(x, value)) {
        const int offset_value = value - min_all_values_;
        // Forward-checking should propagate x != value.
        successor_[x].push_back(offset_value);
      }
    }
    if (successor_[x].size() == 1) {
      const int offset_value = successor_[x][0];
      if (value_to_variable_[offset_value] == -1) {
        value_to_variable_[offset_value] = x;
        variable_to_value_[x] = offset_value;
      }
    }
  }

  // Because we currently propagates all clauses before entering this
  // propagator, we known that this can't happen.
  if (DEBUG_MODE) {
    for (int x = 0; x < num_variables_; x++) {
      for (const int offset_value : successor_[x]) {
        if (value_to_variable_[offset_value] != -1 &&
            value_to_variable_[offset_value] != x) {
          LOG(FATAL) << "Should have been propagated by AllDifferentBinary()!";
        }
      }
    }
  }

  // Seed with previous matching.
  for (int x = 0; x < num_variables_; x++) {
    if (variable_to_value_[x] != -1) continue;
    const int prev_value = prev_matching_[x];
    if (prev_value == -1 || value_to_variable_[prev_value] != -1) continue;

    if (VariableHasPossibleValue(x, prev_matching_[x] + min_all_values_)) {
      variable_to_value_[x] = prev_matching_[x];
      value_to_variable_[prev_matching_[x]] = x;
    }
  }

  // Compute max matching.
  int x = 0;
  for (; x < num_variables_; x++) {
    if (variable_to_value_[x] == -1) {
      value_visited_.assign(num_all_values_, false);
      variable_visited_.assign(num_variables_, false);
      MakeAugmentingPath(x);
    }
    if (variable_to_value_[x] == -1) break;  // No augmenting path exists.
  }

  // Fail if covering variables impossible.
  // Explain with the forbidden parts of the graph that prevent
  // MakeAugmentingPath from increasing the matching size.
  if (x < num_variables_) {
    // For now explain all forbidden arcs.
    std::vector<Literal>* conflict = trail_->MutableConflict();
    conflict->clear();
    for (int y = 0; y < num_variables_; y++) {
      if (!variable_visited_[y]) continue;
      for (int value = variable_min_value_[y]; value <= variable_max_value_[y];
           value++) {
        const LiteralIndex li = VariableLiteralIndexOf(y, value);
        if (li >= 0 && !value_visited_[value - min_all_values_]) {
          DCHECK(trail_->Assignment().LiteralIsFalse(Literal(li)));
          conflict->push_back(Literal(li));
        }
      }
    }
    return false;
  }

  // The current matching is a valid solution, now try to filter values.
  // Build residual graph, compute its SCCs.
  for (int x = 0; x < num_variables_; x++) {
    residual_graph_successors_[x].clear();
    for (const int succ : successor_[x]) {
      if (succ != variable_to_value_[x]) {
        residual_graph_successors_[x].push_back(num_variables_ + succ);
      }
    }
  }
  for (int offset_value = 0; offset_value < num_all_values_; offset_value++) {
    residual_graph_successors_[num_variables_ + offset_value].clear();
    if (value_to_variable_[offset_value] != -1) {
      residual_graph_successors_[num_variables_ + offset_value].push_back(
          value_to_variable_[offset_value]);
    }
  }
  const int dummy_node = num_variables_ + num_all_values_;
  residual_graph_successors_[dummy_node].clear();
  if (num_variables_ < num_all_values_) {
    for (int x = 0; x < num_variables_; x++) {
      residual_graph_successors_[dummy_node].push_back(x);
    }
    for (int offset_value = 0; offset_value < num_all_values_; offset_value++) {
      if (value_to_variable_[offset_value] == -1) {
        residual_graph_successors_[num_variables_ + offset_value].push_back(
            dummy_node);
      }
    }
  }

  // Compute SCCs, make node -> component map.
  struct SccOutput {
    explicit SccOutput(std::vector<int>* c) : components(c) {}
    void emplace_back(int const* b, int const* e) {
      for (int const* it = b; it < e; ++it) {
        (*components)[*it] = num_components;
      }
      ++num_components;
    }
    int num_components = 0;
    std::vector<int>* components;
  };
  SccOutput scc_output(&component_number_);
  FindStronglyConnectedComponents(
      static_cast<int>(residual_graph_successors_.size()),
      residual_graph_successors_, &scc_output);

  // Remove arcs var -> val where SCC(var) -/->* SCC(val).
  for (int x = 0; x < num_variables_; x++) {
    if (successor_[x].size() == 1) continue;
    for (const int offset_value : successor_[x]) {
      const int value_node = offset_value + num_variables_;
      if (variable_to_value_[x] != offset_value &&
          component_number_[x] != component_number_[value_node] &&
          VariableHasPossibleValue(x, offset_value + min_all_values_)) {
        // We can deduce that x != value. To explain, force x == offset_value,
        // then find another assignment for the variable matched to
        // offset_value. It will fail: explaining why is the same as
        // explaining failure as above, and it is an explanation of x != value.
        value_visited_.assign(num_all_values_, false);
        variable_visited_.assign(num_variables_, false);
        // Undo x -> old_value and old_variable -> offset_value.
        const int old_variable = value_to_variable_[offset_value];
        variable_to_value_[old_variable] = -1;
        const int old_value = variable_to_value_[x];
        value_to_variable_[old_value] = -1;
        variable_to_value_[x] = offset_value;
        value_to_variable_[offset_value] = x;

        value_visited_[offset_value] = true;
        MakeAugmentingPath(old_variable);
        DCHECK_EQ(variable_to_value_[old_variable], -1);  // No reassignment.

        std::vector<Literal>* reason = trail_->GetEmptyVectorToStoreReason();
        for (int y = 0; y < num_variables_; y++) {
          if (!variable_visited_[y]) continue;
          for (int value = variable_min_value_[y];
               value <= variable_max_value_[y]; value++) {
            const LiteralIndex li = VariableLiteralIndexOf(y, value);
            if (li >= 0 && !value_visited_[value - min_all_values_]) {
              DCHECK(!VariableHasPossibleValue(y, value));
              reason->push_back(Literal(li));
            }
          }
        }

        const LiteralIndex li =
            VariableLiteralIndexOf(x, offset_value + min_all_values_);
        DCHECK_NE(li, kTrueLiteralIndex);
        DCHECK_NE(li, kFalseLiteralIndex);
        return trail_->EnqueueWithStoredReason(Literal(li).Negated());
      }
    }
  }

  return true;
}

AllDifferentBoundsPropagator::AllDifferentBoundsPropagator(
    const std::vector<AffineExpression>& expressions,
    IntegerTrail* integer_trail)
    : integer_trail_(integer_trail) {
  CHECK(!expressions.empty());

  // We need +2 for sentinels.
  const int capacity = expressions.size() + 2;
  index_to_start_index_.resize(capacity);
  index_to_end_index_.resize(capacity);
  index_is_present_.resize(capacity, false);
  index_to_expr_.resize(capacity, kNoIntegerVariable);

  for (int i = 0; i < expressions.size(); ++i) {
    bounds_.push_back({expressions[i]});
    negated_bounds_.push_back({expressions[i].Negated()});
  }
}

bool AllDifferentBoundsPropagator::Propagate() {
  if (!PropagateLowerBounds()) return false;

  // Note that it is not required to swap back bounds_ and negated_bounds_.
  // TODO(user): investigate the impact.
  std::swap(bounds_, negated_bounds_);
  const bool result = PropagateLowerBounds();
  std::swap(bounds_, negated_bounds_);
  return result;
}

void AllDifferentBoundsPropagator::FillHallReason(IntegerValue hall_lb,
                                                  IntegerValue hall_ub) {
  integer_reason_.clear();
  const int limit = GetIndex(hall_ub);
  for (int i = GetIndex(hall_lb); i <= limit; ++i) {
    const AffineExpression expr = index_to_expr_[i];
    integer_reason_.push_back(expr.GreaterOrEqual(hall_lb));
    integer_reason_.push_back(expr.LowerOrEqual(hall_ub));
  }
}

int AllDifferentBoundsPropagator::FindStartIndexAndCompressPath(int index) {
  // First, walk the pointer until we find one pointing to itself.
  int start_index = index;
  while (true) {
    const int next = index_to_start_index_[start_index];
    if (start_index == next) break;
    start_index = next;
  }

  // Second, redo the same thing and make everyone point to the representative.
  while (true) {
    const int next = index_to_start_index_[index];
    if (start_index == next) break;
    index_to_start_index_[index] = start_index;
    index = next;
  }
  return start_index;
}

bool AllDifferentBoundsPropagator::PropagateLowerBounds() {
  // Start by filling the cached bounds and sorting by increasing lb.
  for (CachedBounds& entry : bounds_) {
    entry.lb = integer_trail_->LowerBound(entry.expr);
    entry.ub = integer_trail_->UpperBound(entry.expr);
  }
  IncrementalSort(bounds_.begin(), bounds_.end(),
                  [](CachedBounds a, CachedBounds b) { return a.lb < b.lb; });

  // We will split the affine epressions in vars sorted by lb in contiguous
  // subset with index of the form [start, start + num_in_window).
  int start = 0;
  int num_in_window = 1;

  // Minimum lower bound in the current window.
  IntegerValue min_lb = bounds_.front().lb;

  const int size = bounds_.size();
  for (int i = 1; i < size; ++i) {
    const IntegerValue lb = bounds_[i].lb;

    // If the lower bounds of all the other variables is greater, then it can
    // never fall into a potential hall interval formed by the variable in the
    // current window, so we can split the problem into independent parts.
    if (lb <= min_lb + IntegerValue(num_in_window - 1)) {
      ++num_in_window;
      continue;
    }

    // Process the current window.
    if (num_in_window > 1) {
      absl::Span<CachedBounds> window(&bounds_[start], num_in_window);
      if (!PropagateLowerBoundsInternal(min_lb, window)) {
        return false;
      }
    }

    // Start of the next window.
    start = i;
    num_in_window = 1;
    min_lb = lb;
  }

  // Take care of the last window.
  if (num_in_window > 1) {
    absl::Span<CachedBounds> window(&bounds_[start], num_in_window);
    return PropagateLowerBoundsInternal(min_lb, window);
  }

  return true;
}

bool AllDifferentBoundsPropagator::PropagateLowerBoundsInternal(
    IntegerValue min_lb, absl::Span<CachedBounds> bounds) {
  hall_starts_.clear();
  hall_ends_.clear();

  // All cached lb in bounds will be in [min_lb, min_lb + bounds_.size()).
  // Make sure we change our base_ so that GetIndex() fit in our buffers.
  base_ = min_lb - IntegerValue(1);

  // Sparse cleaning of index_is_present_.
  for (const int i : indices_to_clear_) {
    index_is_present_[i] = false;
  }
  indices_to_clear_.clear();

  // Sort bounds by increasing ub.
  std::sort(bounds.begin(), bounds.end(),
            [](CachedBounds a, CachedBounds b) { return a.ub < b.ub; });
  for (const CachedBounds entry : bounds) {
    const AffineExpression expr = entry.expr;

    // Note that it is important to use the cache to make sure GetIndex() is
    // not out of bound in case integer_trail_->LowerBound() changed when we
    // pushed something.
    const IntegerValue lb = entry.lb;
    const int lb_index = GetIndex(lb);
    const bool value_is_covered = index_is_present_[lb_index];

    // Check if lb is in an Hall interval, and push it if this is the case.
    if (value_is_covered) {
      const int hall_index =
          std::lower_bound(hall_ends_.begin(), hall_ends_.end(), lb) -
          hall_ends_.begin();
      if (hall_index < hall_ends_.size() && hall_starts_[hall_index] <= lb) {
        const IntegerValue hs = hall_starts_[hall_index];
        const IntegerValue he = hall_ends_[hall_index];
        FillHallReason(hs, he);
        integer_reason_.push_back(expr.GreaterOrEqual(hs));
        if (!integer_trail_->SafeEnqueue(expr.GreaterOrEqual(he + 1),
                                         integer_reason_)) {
          return false;
        }
      }
    }

    // Update our internal representation of the non-consecutive intervals.
    //
    // If lb is not used, we add a node there, otherwise we add it to the
    // right of the interval that contains lb. In both cases, if there is an
    // interval to the left (resp. right) we merge them.
    int new_index = lb_index;
    int start_index = lb_index;
    int end_index = lb_index;
    if (value_is_covered) {
      start_index = FindStartIndexAndCompressPath(new_index);
      new_index = index_to_end_index_[start_index] + 1;
      end_index = new_index;
    } else {
      if (index_is_present_[new_index - 1]) {
        start_index = FindStartIndexAndCompressPath(new_index - 1);
      }
    }
    if (index_is_present_[new_index + 1]) {
      end_index = index_to_end_index_[new_index + 1];
      index_to_start_index_[new_index + 1] = start_index;
    }

    // Update the end of the representative.
    index_to_end_index_[start_index] = end_index;

    // This is the only place where we "add" a new node.
    {
      index_to_start_index_[new_index] = start_index;
      index_to_expr_[new_index] = expr;
      index_is_present_[new_index] = true;
      indices_to_clear_.push_back(new_index);
    }

    // In most situation, we cannot have a conflict now, because it should have
    // been detected before by pushing an interval lower bound past its upper
    // bound. However, it is possible that when we push one bound, other bounds
    // change. So if the upper bound is smaller than the current interval end,
    // we abort so that the conflit reason will be better on the next call to
    // the propagator.
    const IntegerValue end = GetValue(end_index);
    if (end > integer_trail_->UpperBound(expr)) return true;

    // If we have a new Hall interval, add it to the set. Note that it will
    // always be last, and if it overlaps some previous Hall intervals, it
    // always overlaps them fully.
    //
    // Note: It is okay to not use entry.ub here if we want to fetch the last
    // value, but in practice it shouldn't really change when we push a
    // lower_bound and it is faster to use the cached entry.
    if (end == entry.ub) {
      const IntegerValue start = GetValue(start_index);
      while (!hall_starts_.empty() && start <= hall_starts_.back()) {
        hall_starts_.pop_back();
        hall_ends_.pop_back();
      }
      DCHECK(hall_ends_.empty() || hall_ends_.back() < start);
      hall_starts_.push_back(start);
      hall_ends_.push_back(end);
    }
  }
  return true;
}

void AllDifferentBoundsPropagator::RegisterWith(
    GenericLiteralWatcher* watcher) {
  const int id = watcher->Register(this);
  for (const CachedBounds& entry : bounds_) {
    watcher->WatchAffineExpression(entry.expr, id);
  }
  watcher->NotifyThatPropagatorMayNotReachFixedPointInOnePass(id);
}

}  // namespace sat
}  // namespace operations_research