1 //===- Simplex.h - MLIR Simplex Class ---------------------------*- C++ -*-===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // Functionality to perform analysis on FlatAffineConstraints. In particular, 10 // support for performing emptiness checks and redundancy checks. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #ifndef MLIR_ANALYSIS_PRESBURGER_SIMPLEX_H 15 #define MLIR_ANALYSIS_PRESBURGER_SIMPLEX_H 16 17 #include "mlir/Analysis/AffineStructures.h" 18 #include "mlir/Analysis/Presburger/Fraction.h" 19 #include "mlir/Analysis/Presburger/Matrix.h" 20 #include "mlir/Support/LogicalResult.h" 21 #include "llvm/ADT/ArrayRef.h" 22 #include "llvm/ADT/Optional.h" 23 #include "llvm/ADT/SmallVector.h" 24 #include "llvm/Support/raw_ostream.h" 25 26 namespace mlir { 27 28 class GBRSimplex; 29 30 /// This class implements a version of the Simplex and Generalized Basis 31 /// Reduction algorithms, which can perform analysis of integer sets with affine 32 /// inequalities and equalities. A Simplex can be constructed 33 /// by specifying the dimensionality of the set. It supports adding affine 34 /// inequalities and equalities, and can perform emptiness checks, i.e., it can 35 /// find a solution to the set of constraints if one exists, or say that the 36 /// set is empty if no solution exists. Currently, this only works for bounded 37 /// sets. Furthermore, it can find a subset of these constraints that are 38 /// redundant, i.e. a subset of constraints that doesn't constrain the affine 39 /// set further after adding the non-redundant constraints. Simplex can also be 40 /// constructed from a FlatAffineConstraints object. 41 /// 42 /// The implementation of this Simplex class, other than the functionality 43 /// for sampling, is based on the paper 44 /// "Simplify: A Theorem Prover for Program Checking" 45 /// by D. Detlefs, G. Nelson, J. B. Saxe. 46 /// 47 /// We define variables, constraints, and unknowns. Consider the example of a 48 /// two-dimensional set defined by 1 + 2x + 3y >= 0 and 2x - 3y >= 0. Here, 49 /// x, y, are variables while 1 + 2x + 3y >= 0, 2x - 3y >= 0 are 50 /// constraints. Unknowns are either variables or constraints, i.e., x, y, 51 /// 1 + 2x + 3y >= 0, 2x - 3y >= 0 are all unknowns. 52 /// 53 /// The implementation involves a matrix called a tableau, which can be thought 54 /// of as a 2D matrix of rational numbers having number of rows equal to the 55 /// number of constraints and number of columns equal to one plus the number of 56 /// variables. In our implementation, instead of storing rational numbers, we 57 /// store a common denominator for each row, so it is in fact a matrix of 58 /// integers with number of rows equal to number of constraints and number of 59 /// columns equal to _two_ plus the number of variables. For example, instead of 60 /// storing a row of three rationals [1/2, 2/3, 3], we would store [6, 3, 4, 18] 61 /// since 3/6 = 1/2, 4/6 = 2/3, and 18/6 = 3. 62 /// 63 /// Every row and column except the first and second columns is associated with 64 /// an unknown and every unknown is associated with a row or column. The second 65 /// column represents the constant, explained in more detail below. An unknown 66 /// associated with a row or column is said to be in row or column position 67 /// respectively. 68 /// 69 /// The vectors var and con store information about the variables and 70 /// constraints respectively, namely, whether they are in row or column 71 /// position, which row or column they are associated with, and whether they 72 /// correspond to a variable or a constraint. 73 /// 74 /// An unknown is addressed by its index. If the index i is non-negative, then 75 /// the variable var[i] is being addressed. If the index i is negative, then 76 /// the constraint con[~i] is being addressed. Effectively this maps 77 /// 0 -> var[0], 1 -> var[1], -1 -> con[0], -2 -> con[1], etc. rowUnknown[r] and 78 /// colUnknown[c] are the indexes of the unknowns associated with row r and 79 /// column c, respectively. 80 /// 81 /// The unknowns in column position are together called the basis. Initially the 82 /// basis is the set of variables -- in our example above, the initial basis is 83 /// x, y. 84 /// 85 /// The unknowns in row position are represented in terms of the basis unknowns. 86 /// If the basis unknowns are u_1, u_2, ... u_m, and a row in the tableau is 87 /// d, c, a_1, a_2, ... a_m, this representats the unknown for that row as 88 /// (c + a_1*u_1 + a_2*u_2 + ... + a_m*u_m)/d. In our running example, if the 89 /// basis is the initial basis of x, y, then the constraint 1 + 2x + 3y >= 0 90 /// would be represented by the row [1, 1, 2, 3]. 91 /// 92 /// The association of unknowns to rows and columns can be changed by a process 93 /// called pivoting, where a row unknown and a column unknown exchange places 94 /// and the remaining row variables' representation is changed accordingly 95 /// by eliminating the old column unknown in favour of the new column unknown. 96 /// If we had pivoted the column for x with the row for 2x - 3y >= 0, 97 /// the new row for x would be [2, 1, 3] since x = (1*(2x - 3y) + 3*y)/2. 98 /// See the documentation for the pivot member function for details. 99 /// 100 /// The association of unknowns to rows and columns is called the _tableau 101 /// configuration_. The _sample value_ of an unknown in a particular tableau 102 /// configuration is its value if all the column unknowns were set to zero. 103 /// Concretely, for unknowns in column position the sample value is zero and 104 /// for unknowns in row position the sample value is the constant term divided 105 /// by the common denominator. 106 /// 107 /// The tableau configuration is called _consistent_ if the sample value of all 108 /// restricted unknowns is non-negative. Initially there are no constraints, and 109 /// the tableau is consistent. When a new constraint is added, its sample value 110 /// in the current tableau configuration may be negative. In that case, we try 111 /// to find a series of pivots to bring us to a consistent tableau 112 /// configuration, i.e. we try to make the new constraint's sample value 113 /// non-negative without making that of any other constraints negative. (See 114 /// findPivot and findPivotRow for details.) If this is not possible, then the 115 /// set of constraints is mutually contradictory and the tableau is marked 116 /// _empty_, which means the set of constraints has no solution. 117 /// 118 /// The Simplex class supports redundancy checking via detectRedundant and 119 /// isMarkedRedundant. A redundant constraint is one which is never violated as 120 /// long as the other constraints are not violated, i.e., removing a redundant 121 /// constraint does not change the set of solutions to the constraints. As a 122 /// heuristic, constraints that have been marked redundant can be ignored for 123 /// most operations. Therefore, these constraints are kept in rows 0 to 124 /// nRedundant - 1, where nRedundant is a member variable that tracks the number 125 /// of constraints that have been marked redundant. 126 /// 127 /// This Simplex class also supports taking snapshots of the current state 128 /// and rolling back to prior snapshots. This works by maintaining an undo log 129 /// of operations. Snapshots are just pointers to a particular location in the 130 /// log, and rolling back to a snapshot is done by reverting each log entry's 131 /// operation from the end until we reach the snapshot's location. 132 /// 133 /// Finding an integer sample is done with the Generalized Basis Reduction 134 /// algorithm. See the documentation for findIntegerSample and reduceBasis. 135 class Simplex { 136 public: 137 enum class Direction { Up, Down }; 138 139 Simplex() = delete; 140 explicit Simplex(unsigned nVar); 141 explicit Simplex(const FlatAffineConstraints &constraints); 142 143 /// Returns true if the tableau is empty (has conflicting constraints), 144 /// false otherwise. 145 bool isEmpty() const; 146 147 /// Add an inequality to the tableau. If coeffs is c_0, c_1, ... c_n, where n 148 /// is the current number of variables, then the corresponding inequality is 149 /// c_n + c_0*x_0 + c_1*x_1 + ... + c_{n-1}*x_{n-1} >= 0. 150 void addInequality(ArrayRef<int64_t> coeffs); 151 152 /// Returns the number of variables in the tableau. 153 unsigned numVariables() const; 154 155 /// Returns the number of constraints in the tableau. 156 unsigned numConstraints() const; 157 158 /// Add an equality to the tableau. If coeffs is c_0, c_1, ... c_n, where n 159 /// is the current number of variables, then the corresponding equality is 160 /// c_n + c_0*x_0 + c_1*x_1 + ... + c_{n-1}*x_{n-1} == 0. 161 void addEquality(ArrayRef<int64_t> coeffs); 162 163 /// Mark the tableau as being empty. 164 void markEmpty(); 165 166 /// Get a snapshot of the current state. This is used for rolling back. 167 unsigned getSnapshot() const; 168 169 /// Rollback to a snapshot. This invalidates all later snapshots. 170 void rollback(unsigned snapshot); 171 172 /// Add all the constraints from the given FlatAffineConstraints. 173 void intersectFlatAffineConstraints(const FlatAffineConstraints &fac); 174 175 /// Compute the maximum or minimum value of the given row, depending on 176 /// direction. The specified row is never pivoted. 177 /// 178 /// Returns a (num, den) pair denoting the optimum, or None if no 179 /// optimum exists, i.e., if the expression is unbounded in this direction. 180 Optional<Fraction> computeRowOptimum(Direction direction, unsigned row); 181 182 /// Compute the maximum or minimum value of the given expression, depending on 183 /// direction. 184 /// 185 /// Returns a (num, den) pair denoting the optimum, or a null value if no 186 /// optimum exists, i.e., if the expression is unbounded in this direction. 187 Optional<Fraction> computeOptimum(Direction direction, 188 ArrayRef<int64_t> coeffs); 189 190 /// Returns whether the specified constraint has been marked as redundant. 191 /// Constraints are numbered from 0 starting at the first added inequality. 192 /// Equalities are added as a pair of inequalities and so correspond to two 193 /// inequalities with successive indices. 194 bool isMarkedRedundant(unsigned constraintIndex) const; 195 196 /// Finds a subset of constraints that is redundant, i.e., such that 197 /// the set of solutions does not change if these constraints are removed. 198 /// Marks these constraints as redundant. Whether a specific constraint has 199 /// been marked redundant can be queried using isMarkedRedundant. 200 void detectRedundant(); 201 202 /// Returns a (min, max) pair denoting the minimum and maximum integer values 203 /// of the given expression. 204 std::pair<int64_t, int64_t> computeIntegerBounds(ArrayRef<int64_t> coeffs); 205 206 /// Returns true if the polytope is unbounded, i.e., extends to infinity in 207 /// some direction. Otherwise, returns false. 208 bool isUnbounded(); 209 210 /// Make a tableau to represent a pair of points in the given tableaus, one in 211 /// tableau A and one in B. 212 static Simplex makeProduct(const Simplex &a, const Simplex &b); 213 214 /// Returns the current sample point if it is integral. Otherwise, returns an 215 /// None. 216 Optional<SmallVector<int64_t, 8>> getSamplePointIfIntegral() const; 217 218 /// Returns an integer sample point if one exists, or None 219 /// otherwise. This should only be called for bounded sets. 220 Optional<SmallVector<int64_t, 8>> findIntegerSample(); 221 222 /// Print the tableau's internal state. 223 void print(raw_ostream &os) const; 224 void dump() const; 225 226 private: 227 friend class GBRSimplex; 228 229 enum class Orientation { Row, Column }; 230 231 /// An Unknown is either a variable or a constraint. It is always associated 232 /// with either a row or column. Whether it's a row or a column is specified 233 /// by the orientation and pos identifies the specific row or column it is 234 /// associated with. If the unknown is restricted, then it has a 235 /// non-negativity constraint associated with it, i.e., its sample value must 236 /// always be non-negative and if it cannot be made non-negative without 237 /// violating other constraints, the tableau is empty. 238 struct Unknown { UnknownUnknown239 Unknown(Orientation oOrientation, bool oRestricted, unsigned oPos) 240 : pos(oPos), orientation(oOrientation), restricted(oRestricted) {} 241 unsigned pos; 242 Orientation orientation; 243 bool restricted : 1; 244 printUnknown245 void print(raw_ostream &os) const { 246 os << (orientation == Orientation::Row ? "r" : "c"); 247 os << pos; 248 if (restricted) 249 os << " [>=0]"; 250 } 251 }; 252 253 struct Pivot { 254 unsigned row, column; 255 }; 256 257 /// Find a pivot to change the sample value of row in the specified 258 /// direction. The returned pivot row will be row if and only 259 /// if the unknown is unbounded in the specified direction. 260 /// 261 /// Returns a (row, col) pair denoting a pivot, or an empty Optional if 262 /// no valid pivot exists. 263 Optional<Pivot> findPivot(int row, Direction direction) const; 264 265 /// Swap the row with the column in the tableau's data structures but not the 266 /// tableau itself. This is used by pivot. 267 void swapRowWithCol(unsigned row, unsigned col); 268 269 /// Pivot the row with the column. 270 void pivot(unsigned row, unsigned col); 271 void pivot(Pivot pair); 272 273 /// Returns the unknown associated with index. 274 const Unknown &unknownFromIndex(int index) const; 275 /// Returns the unknown associated with col. 276 const Unknown &unknownFromColumn(unsigned col) const; 277 /// Returns the unknown associated with row. 278 const Unknown &unknownFromRow(unsigned row) const; 279 /// Returns the unknown associated with index. 280 Unknown &unknownFromIndex(int index); 281 /// Returns the unknown associated with col. 282 Unknown &unknownFromColumn(unsigned col); 283 /// Returns the unknown associated with row. 284 Unknown &unknownFromRow(unsigned row); 285 286 /// Add a new row to the tableau and the associated data structures. 287 unsigned addRow(ArrayRef<int64_t> coeffs); 288 289 /// Normalize the given row by removing common factors between the numerator 290 /// and the denominator. 291 void normalizeRow(unsigned row); 292 293 /// Swap the two rows in the tableau and associated data structures. 294 void swapRows(unsigned i, unsigned j); 295 296 /// Restore the unknown to a non-negative sample value. 297 /// 298 /// Returns true if the unknown was successfully restored to a non-negative 299 /// sample value, false otherwise. 300 LogicalResult restoreRow(Unknown &u); 301 302 /// Mark the specified unknown redundant. This operation is added to the undo 303 /// log and will be undone by rollbacks. The specified unknown must be in row 304 /// orientation. 305 void markRowRedundant(Unknown &u); 306 307 /// Enum to denote operations that need to be undone during rollback. 308 enum class UndoLogEntry { 309 RemoveLastConstraint, 310 UnmarkEmpty, 311 UnmarkLastRedundant 312 }; 313 314 /// Undo the operation represented by the log entry. 315 void undo(UndoLogEntry entry); 316 317 /// Find a row that can be used to pivot the column in the specified 318 /// direction. If skipRow is not null, then this row is excluded 319 /// from consideration. The returned pivot will maintain all constraints 320 /// except the column itself and skipRow, if it is set. (if these unknowns 321 /// are restricted). 322 /// 323 /// Returns the row to pivot to, or an empty Optional if the column 324 /// is unbounded in the specified direction. 325 Optional<unsigned> findPivotRow(Optional<unsigned> skipRow, 326 Direction direction, unsigned col) const; 327 328 /// Reduce the given basis, starting at the specified level, using general 329 /// basis reduction. 330 void reduceBasis(Matrix &basis, unsigned level); 331 332 /// The number of rows in the tableau. 333 unsigned nRow; 334 335 /// The number of columns in the tableau, including the common denominator 336 /// and the constant column. 337 unsigned nCol; 338 339 /// The number of redundant rows in the tableau. These are the first 340 /// nRedundant rows. 341 unsigned nRedundant; 342 343 /// The matrix representing the tableau. 344 Matrix tableau; 345 346 /// This is true if the tableau has been detected to be empty, false 347 /// otherwise. 348 bool empty; 349 350 /// Holds a log of operations, used for rolling back to a previous state. 351 SmallVector<UndoLogEntry, 8> undoLog; 352 353 /// These hold the indexes of the unknown at a given row or column position. 354 /// We keep these as signed integers since that makes it convenient to check 355 /// if an index corresponds to a variable or a constraint by checking the 356 /// sign. 357 /// 358 /// colUnknown is padded with two null indexes at the front since the first 359 /// two columns don't correspond to any unknowns. 360 SmallVector<int, 8> rowUnknown, colUnknown; 361 362 /// These hold information about each unknown. 363 SmallVector<Unknown, 8> con, var; 364 }; 365 366 } // namespace mlir 367 368 #endif // MLIR_ANALYSIS_PRESBURGER_SIMPLEX_H 369