GCC Code Coverage Report
Directory: . Exec Total Coverage
File: src/theory/arith/fc_simplex.h Lines: 2 39 5.1 %
Date: 2021-05-22 Branches: 0 26 0.0 %

Line Exec Source
1
/******************************************************************************
2
 * Top contributors (to current version):
3
 *   Tim King, Morgan Deters, Mathias Preiner
4
 *
5
 * This file is part of the cvc5 project.
6
 *
7
 * Copyright (c) 2009-2021 by the authors listed in the file AUTHORS
8
 * in the top-level source directory and their institutional affiliations.
9
 * All rights reserved.  See the file COPYING in the top-level source
10
 * directory for licensing information.
11
 * ****************************************************************************
12
 *
13
 * This is an implementation of the Simplex Module for the Simplex for
14
 * DPLL(T)decision procedure.
15
 *
16
 * This implements the Simplex module for the Simpelx for DPLL(T) decision
17
 * procedure.
18
 * See the Simplex for DPLL(T) technical report for more background.(citation?)
19
 * This shares with the theory a Tableau, and a PartialModel that:
20
 *  - satisfies the equalities in the Tableau, and
21
 *  - the assignment for the non-basic variables satisfies their bounds.
22
 * This is required to either produce a conflict or satisifying PartialModel.
23
 * Further, we require being told when a basic variable updates its value.
24
 *
25
 * During the Simplex search we maintain a queue of variables.
26
 * The queue is required to contain all of the basic variables that voilate
27
 * their bounds.
28
 * As elimination from the queue is more efficient to be done lazily,
29
 * we do not maintain that the queue of variables needs to be only basic
30
 * variables or only variables that satisfy their bounds.
31
 *
32
 * The simplex procedure roughly follows Alberto's thesis. (citation?)
33
 * There is one round of selecting using a heuristic pivoting rule.
34
 * (See PreferenceFunction Documentation for the available options.)
35
 * The non-basic variable is the one that appears in the fewest pivots.
36
 * (Bruno says that Leonardo invented this first.)
37
 * After this, Bland's pivot rule is invoked.
38
 *
39
 * During this proccess, we periodically inspect the queue of variables to
40
 * 1) remove now extraneous extries,
41
 * 2) detect conflicts that are "waiting" on the queue but may not be detected
42
 *    by the current queue heuristics, and
43
 * 3) detect multiple conflicts.
44
 *
45
 * Conflicts are greedily slackened to use the weakest bounds that still
46
 * produce the conflict.
47
 *
48
 * Extra things tracked atm: (Subject to change at Tim's whims)
49
 * - A superset of all of the newly pivoted variables.
50
 * - A queue of additional conflicts that were discovered by Simplex.
51
 *   These are theory valid and are currently turned into lemmas
52
 */
53
54
#include "cvc5_private.h"
55
56
#pragma once
57
58
#include "theory/arith/error_set.h"
59
#include "theory/arith/linear_equality.h"
60
#include "theory/arith/simplex.h"
61
#include "theory/arith/simplex_update.h"
62
#include "util/dense_map.h"
63
#include "util/statistics_stats.h"
64
65
namespace cvc5 {
66
namespace theory {
67
namespace arith {
68
69
9459
class FCSimplexDecisionProcedure : public SimplexDecisionProcedure{
70
public:
71
  FCSimplexDecisionProcedure(LinearEqualityModule& linEq, ErrorSet& errors, RaiseConflict conflictChannel, TempVarMalloc tvmalloc);
72
73
  Result::Sat findModel(bool exactResult) override;
74
75
  // other error variables are dropping
76
  WitnessImprovement dualLikeImproveError(ArithVar evar);
77
  WitnessImprovement primalImproveError(ArithVar evar);
78
79
  // dual like
80
  // - found conflict
81
  // - satisfied error set
82
  Result::Sat dualLike();
83
84
private:
85
  static const uint32_t PENALTY = 4;
86
  DenseMultiset d_scores;
87
  void decreasePenalties(){ d_scores.removeOneOfEverything(); }
88
  uint32_t penalty(ArithVar x) const { return d_scores.count(x); }
89
  void setPenalty(ArithVar x, WitnessImprovement w){
90
    if(improvement(w)){
91
      if(d_scores.count(x) > 0){
92
        d_scores.removeAll(x);
93
      }
94
    }else{
95
      d_scores.setCount(x, PENALTY);
96
    }
97
  }
98
99
  /** The size of the focus set. */
100
  uint32_t d_focusSize;
101
102
  /** The current error focus variable. */
103
  ArithVar d_focusErrorVar;
104
105
  /**
106
   * The signs of the coefficients in the focus set.
107
   * This is empty until this has been loaded.
108
   */
109
  DenseMap<const Rational*> d_focusCoefficients;
110
111
  /**
112
   * Loads the signs of the coefficients of the variables on the row d_focusErrorVar
113
   * into d_focusSgns.
114
   */
115
  void loadFocusSigns();
116
117
  /** Unloads the information from d_focusSgns. */
118
  void unloadFocusSigns();
119
120
  /**
121
   * The signs of a variable in the row of d_focusErrorVar.
122
   * d_focusSgns must be loaded.
123
   */
124
  const Rational& focusCoefficient(ArithVar nb) const;
125
126
  int32_t d_pivotBudget;
127
  // enum PivotImprovement {
128
  //   ErrorDropped,
129
  //   NonDegenerate,
130
  //   HeuristicDegenerate,
131
  //   BlandsDegenerate
132
  // };
133
134
  WitnessImprovement d_prevWitnessImprovement;
135
  uint32_t d_witnessImprovementInARow;
136
137
  uint32_t degeneratePivotsInARow() const;
138
139
  static const uint32_t s_focusThreshold = 6;
140
  static const uint32_t s_maxDegeneratePivotsBeforeBlandsOnLeaving = 100;
141
  static const uint32_t s_maxDegeneratePivotsBeforeBlandsOnEntering = 10;
142
143
  DenseMap<uint32_t> d_leavingCountSinceImprovement;
144
  void increaseLeavingCount(ArithVar x){
145
    if(!d_leavingCountSinceImprovement.isKey(x)){
146
      d_leavingCountSinceImprovement.set(x,1);
147
    }else{
148
      (d_leavingCountSinceImprovement.get(x))++;
149
    }
150
  }
151
  LinearEqualityModule::UpdatePreferenceFunction selectLeavingFunction(ArithVar x){
152
    bool useBlands = d_leavingCountSinceImprovement.isKey(x) &&
153
      d_leavingCountSinceImprovement[x] >= s_maxDegeneratePivotsBeforeBlandsOnEntering;
154
    if(useBlands) {
155
      return &LinearEqualityModule::preferWitness<false>;
156
    } else {
157
      return &LinearEqualityModule::preferWitness<true>;
158
    }
159
  }
160
161
  bool debugDualLike(WitnessImprovement w, std::ostream& out,
162
                     int instance,
163
                     uint32_t prevFocusSize, uint32_t prevErrorSize) const;
164
165
  void debugPrintSignal(ArithVar updated) const;
166
167
  ArithVarVec d_sgnDisagreements;
168
169
  //static PivotImprovement pivotImprovement(const UpdateInfo& selected, bool useBlands = false);
170
171
  void logPivot(WitnessImprovement w);
172
173
  void updateAndSignal(const UpdateInfo& selected, WitnessImprovement w);
174
175
  UpdateInfo selectPrimalUpdate(ArithVar error,
176
                                LinearEqualityModule::UpdatePreferenceFunction upf,
177
                                LinearEqualityModule::VarPreferenceFunction bpf);
178
179
180
  UpdateInfo selectUpdateForDualLike(ArithVar basic){
181
    TimerStat::CodeTimer codeTimer(d_statistics.d_selectUpdateForDualLike);
182
183
    LinearEqualityModule::UpdatePreferenceFunction upf =
184
      &LinearEqualityModule::preferWitness<true>;
185
    LinearEqualityModule::VarPreferenceFunction bpf =
186
      &LinearEqualityModule::minVarOrder;
187
    return selectPrimalUpdate(basic, upf, bpf);
188
  }
189
190
  UpdateInfo selectUpdateForPrimal(ArithVar basic, bool useBlands){
191
    TimerStat::CodeTimer codeTimer(d_statistics.d_selectUpdateForPrimal);
192
193
    LinearEqualityModule::UpdatePreferenceFunction upf;
194
    if(useBlands) {
195
      upf = &LinearEqualityModule::preferWitness<false>;
196
    } else {
197
      upf = &LinearEqualityModule::preferWitness<true>;
198
    }
199
200
    LinearEqualityModule::VarPreferenceFunction bpf = useBlands ?
201
      &LinearEqualityModule::minVarOrder :
202
      &LinearEqualityModule::minRowLength;
203
204
    return selectPrimalUpdate(basic, upf, bpf);
205
  }
206
  WitnessImprovement selectFocusImproving() ;
207
208
  WitnessImprovement focusUsingSignDisagreements(ArithVar basic);
209
  WitnessImprovement focusDownToLastHalf();
210
  WitnessImprovement adjustFocusShrank(const ArithVarVec& drop);
211
  WitnessImprovement focusDownToJust(ArithVar v);
212
213
214
  void adjustFocusAndError(const UpdateInfo& up, const AVIntPairVec& focusChanges);
215
216
  /**
217
   * This is the main simplex for DPLL(T) loop.
218
   * It runs for at most maxIterations.
219
   *
220
   * Returns true iff it has found a conflict.
221
   * d_conflictVariable will be set and the conflict for this row is reported.
222
   */
223
  bool searchForFeasibleSolution(uint32_t maxIterations);
224
225
  bool initialProcessSignals(){
226
    TimerStat &timer = d_statistics.d_initialSignalsTime;
227
    IntStat& conflictStat  = d_statistics.d_initialConflicts;
228
    bool res = standardProcessSignals(timer, conflictStat);
229
    d_focusSize = d_errorSet.focusSize();
230
    return res;
231
  }
232
233
  static bool debugCheckWitness(const UpdateInfo& inf, WitnessImprovement w, bool useBlands);
234
235
  /** These fields are designed to be accessible to TheoryArith methods. */
236
9459
  class Statistics {
237
  public:
238
    TimerStat d_initialSignalsTime;
239
    IntStat d_initialConflicts;
240
241
    IntStat d_fcFoundUnsat;
242
    IntStat d_fcFoundSat;
243
    IntStat d_fcMissed;
244
245
    TimerStat d_fcTimer;
246
    TimerStat d_fcFocusConstructionTimer;
247
248
    TimerStat d_selectUpdateForDualLike;
249
    TimerStat d_selectUpdateForPrimal;
250
251
    ReferenceStat<uint32_t> d_finalCheckPivotCounter;
252
253
    Statistics(const std::string& name, uint32_t& pivots);
254
  } d_statistics;
255
};/* class FCSimplexDecisionProcedure */
256
257
}  // namespace arith
258
}  // namespace theory
259
}  // namespace cvc5