Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 1 | //===- llvm/ADT/BitVector.h - Bit vectors -----------------------*- C++ -*-===// |
| 2 | // |
| 3 | // The LLVM Compiler Infrastructure |
| 4 | // |
| 5 | // This file is distributed under the University of Illinois Open Source |
| 6 | // License. See LICENSE.TXT for details. |
| 7 | // |
| 8 | //===----------------------------------------------------------------------===// |
| 9 | // |
| 10 | // This file implements the BitVector class. |
| 11 | // |
| 12 | //===----------------------------------------------------------------------===// |
| 13 | |
| 14 | #ifndef LLVM_ADT_BITVECTOR_H |
| 15 | #define LLVM_ADT_BITVECTOR_H |
| 16 | |
| 17 | #include "llvm/ADT/ArrayRef.h" |
| 18 | #include "llvm/ADT/iterator_range.h" |
| 19 | #include "llvm/Support/MathExtras.h" |
| 20 | #include <algorithm> |
| 21 | #include <cassert> |
| 22 | #include <climits> |
| 23 | #include <cstdint> |
| 24 | #include <cstdlib> |
| 25 | #include <cstring> |
| 26 | #include <utility> |
| 27 | |
| 28 | namespace llvm { |
| 29 | |
| 30 | /// ForwardIterator for the bits that are set. |
| 31 | /// Iterators get invalidated when resize / reserve is called. |
| 32 | template <typename BitVectorT> class const_set_bits_iterator_impl { |
| 33 | const BitVectorT &Parent; |
| 34 | int Current = 0; |
| 35 | |
| 36 | void advance() { |
| 37 | assert(Current != -1 && "Trying to advance past end."); |
| 38 | Current = Parent.find_next(Current); |
| 39 | } |
| 40 | |
| 41 | public: |
| 42 | const_set_bits_iterator_impl(const BitVectorT &Parent, int Current) |
| 43 | : Parent(Parent), Current(Current) {} |
| 44 | explicit const_set_bits_iterator_impl(const BitVectorT &Parent) |
| 45 | : const_set_bits_iterator_impl(Parent, Parent.find_first()) {} |
| 46 | const_set_bits_iterator_impl(const const_set_bits_iterator_impl &) = default; |
| 47 | |
| 48 | const_set_bits_iterator_impl operator++(int) { |
| 49 | auto Prev = *this; |
| 50 | advance(); |
| 51 | return Prev; |
| 52 | } |
| 53 | |
| 54 | const_set_bits_iterator_impl &operator++() { |
| 55 | advance(); |
| 56 | return *this; |
| 57 | } |
| 58 | |
| 59 | unsigned operator*() const { return Current; } |
| 60 | |
| 61 | bool operator==(const const_set_bits_iterator_impl &Other) const { |
| 62 | assert(&Parent == &Other.Parent && |
| 63 | "Comparing iterators from different BitVectors"); |
| 64 | return Current == Other.Current; |
| 65 | } |
| 66 | |
| 67 | bool operator!=(const const_set_bits_iterator_impl &Other) const { |
| 68 | assert(&Parent == &Other.Parent && |
| 69 | "Comparing iterators from different BitVectors"); |
| 70 | return Current != Other.Current; |
| 71 | } |
| 72 | }; |
| 73 | |
| 74 | class BitVector { |
| 75 | typedef unsigned long BitWord; |
| 76 | |
| 77 | enum { BITWORD_SIZE = (unsigned)sizeof(BitWord) * CHAR_BIT }; |
| 78 | |
| 79 | static_assert(BITWORD_SIZE == 64 || BITWORD_SIZE == 32, |
| 80 | "Unsupported word size"); |
| 81 | |
| 82 | MutableArrayRef<BitWord> Bits; // Actual bits. |
| 83 | unsigned Size; // Size of bitvector in bits. |
| 84 | |
| 85 | public: |
| 86 | typedef unsigned size_type; |
| 87 | // Encapsulation of a single bit. |
| 88 | class reference { |
| 89 | friend class BitVector; |
| 90 | |
| 91 | BitWord *WordRef; |
| 92 | unsigned BitPos; |
| 93 | |
| 94 | public: |
| 95 | reference(BitVector &b, unsigned Idx) { |
| 96 | WordRef = &b.Bits[Idx / BITWORD_SIZE]; |
| 97 | BitPos = Idx % BITWORD_SIZE; |
| 98 | } |
| 99 | |
| 100 | reference() = delete; |
| 101 | reference(const reference&) = default; |
| 102 | |
| 103 | reference &operator=(reference t) { |
| 104 | *this = bool(t); |
| 105 | return *this; |
| 106 | } |
| 107 | |
| 108 | reference& operator=(bool t) { |
| 109 | if (t) |
| 110 | *WordRef |= BitWord(1) << BitPos; |
| 111 | else |
| 112 | *WordRef &= ~(BitWord(1) << BitPos); |
| 113 | return *this; |
| 114 | } |
| 115 | |
| 116 | operator bool() const { |
| 117 | return ((*WordRef) & (BitWord(1) << BitPos)) != 0; |
| 118 | } |
| 119 | }; |
| 120 | |
| 121 | typedef const_set_bits_iterator_impl<BitVector> const_set_bits_iterator; |
| 122 | typedef const_set_bits_iterator set_iterator; |
| 123 | |
| 124 | const_set_bits_iterator set_bits_begin() const { |
| 125 | return const_set_bits_iterator(*this); |
| 126 | } |
| 127 | const_set_bits_iterator set_bits_end() const { |
| 128 | return const_set_bits_iterator(*this, -1); |
| 129 | } |
| 130 | iterator_range<const_set_bits_iterator> set_bits() const { |
| 131 | return make_range(set_bits_begin(), set_bits_end()); |
| 132 | } |
| 133 | |
| 134 | /// BitVector default ctor - Creates an empty bitvector. |
| 135 | BitVector() : Size(0) {} |
| 136 | |
| 137 | /// BitVector ctor - Creates a bitvector of specified number of bits. All |
| 138 | /// bits are initialized to the specified value. |
| 139 | explicit BitVector(unsigned s, bool t = false) : Size(s) { |
| 140 | size_t Capacity = NumBitWords(s); |
| 141 | Bits = allocate(Capacity); |
| 142 | init_words(Bits, t); |
| 143 | if (t) |
| 144 | clear_unused_bits(); |
| 145 | } |
| 146 | |
| 147 | /// BitVector copy ctor. |
| 148 | BitVector(const BitVector &RHS) : Size(RHS.size()) { |
| 149 | if (Size == 0) { |
| 150 | Bits = MutableArrayRef<BitWord>(); |
| 151 | return; |
| 152 | } |
| 153 | |
| 154 | size_t Capacity = NumBitWords(RHS.size()); |
| 155 | Bits = allocate(Capacity); |
| 156 | std::memcpy(Bits.data(), RHS.Bits.data(), Capacity * sizeof(BitWord)); |
| 157 | } |
| 158 | |
| 159 | BitVector(BitVector &&RHS) : Bits(RHS.Bits), Size(RHS.Size) { |
| 160 | RHS.Bits = MutableArrayRef<BitWord>(); |
| 161 | RHS.Size = 0; |
| 162 | } |
| 163 | |
| 164 | ~BitVector() { std::free(Bits.data()); } |
| 165 | |
| 166 | /// empty - Tests whether there are no bits in this bitvector. |
| 167 | bool empty() const { return Size == 0; } |
| 168 | |
| 169 | /// size - Returns the number of bits in this bitvector. |
| 170 | size_type size() const { return Size; } |
| 171 | |
| 172 | /// count - Returns the number of bits which are set. |
| 173 | size_type count() const { |
| 174 | unsigned NumBits = 0; |
| 175 | for (unsigned i = 0; i < NumBitWords(size()); ++i) |
| 176 | NumBits += countPopulation(Bits[i]); |
| 177 | return NumBits; |
| 178 | } |
| 179 | |
| 180 | /// any - Returns true if any bit is set. |
| 181 | bool any() const { |
| 182 | for (unsigned i = 0; i < NumBitWords(size()); ++i) |
| 183 | if (Bits[i] != 0) |
| 184 | return true; |
| 185 | return false; |
| 186 | } |
| 187 | |
| 188 | /// all - Returns true if all bits are set. |
| 189 | bool all() const { |
| 190 | for (unsigned i = 0; i < Size / BITWORD_SIZE; ++i) |
| 191 | if (Bits[i] != ~0UL) |
| 192 | return false; |
| 193 | |
| 194 | // If bits remain check that they are ones. The unused bits are always zero. |
| 195 | if (unsigned Remainder = Size % BITWORD_SIZE) |
| 196 | return Bits[Size / BITWORD_SIZE] == (1UL << Remainder) - 1; |
| 197 | |
| 198 | return true; |
| 199 | } |
| 200 | |
| 201 | /// none - Returns true if none of the bits are set. |
| 202 | bool none() const { |
| 203 | return !any(); |
| 204 | } |
| 205 | |
| 206 | /// find_first_in - Returns the index of the first set bit in the range |
| 207 | /// [Begin, End). Returns -1 if all bits in the range are unset. |
| 208 | int find_first_in(unsigned Begin, unsigned End) const { |
| 209 | assert(Begin <= End && End <= Size); |
| 210 | if (Begin == End) |
| 211 | return -1; |
| 212 | |
| 213 | unsigned FirstWord = Begin / BITWORD_SIZE; |
| 214 | unsigned LastWord = (End - 1) / BITWORD_SIZE; |
| 215 | |
| 216 | // Check subsequent words. |
| 217 | for (unsigned i = FirstWord; i <= LastWord; ++i) { |
| 218 | BitWord Copy = Bits[i]; |
| 219 | |
| 220 | if (i == FirstWord) { |
| 221 | unsigned FirstBit = Begin % BITWORD_SIZE; |
| 222 | Copy &= maskTrailingZeros<BitWord>(FirstBit); |
| 223 | } |
| 224 | |
| 225 | if (i == LastWord) { |
| 226 | unsigned LastBit = (End - 1) % BITWORD_SIZE; |
| 227 | Copy &= maskTrailingOnes<BitWord>(LastBit + 1); |
| 228 | } |
| 229 | if (Copy != 0) |
| 230 | return i * BITWORD_SIZE + countTrailingZeros(Copy); |
| 231 | } |
| 232 | return -1; |
| 233 | } |
| 234 | |
| 235 | /// find_last_in - Returns the index of the last set bit in the range |
| 236 | /// [Begin, End). Returns -1 if all bits in the range are unset. |
| 237 | int find_last_in(unsigned Begin, unsigned End) const { |
| 238 | assert(Begin <= End && End <= Size); |
| 239 | if (Begin == End) |
| 240 | return -1; |
| 241 | |
| 242 | unsigned LastWord = (End - 1) / BITWORD_SIZE; |
| 243 | unsigned FirstWord = Begin / BITWORD_SIZE; |
| 244 | |
| 245 | for (unsigned i = LastWord + 1; i >= FirstWord + 1; --i) { |
| 246 | unsigned CurrentWord = i - 1; |
| 247 | |
| 248 | BitWord Copy = Bits[CurrentWord]; |
| 249 | if (CurrentWord == LastWord) { |
| 250 | unsigned LastBit = (End - 1) % BITWORD_SIZE; |
| 251 | Copy &= maskTrailingOnes<BitWord>(LastBit + 1); |
| 252 | } |
| 253 | |
| 254 | if (CurrentWord == FirstWord) { |
| 255 | unsigned FirstBit = Begin % BITWORD_SIZE; |
| 256 | Copy &= maskTrailingZeros<BitWord>(FirstBit); |
| 257 | } |
| 258 | |
| 259 | if (Copy != 0) |
| 260 | return (CurrentWord + 1) * BITWORD_SIZE - countLeadingZeros(Copy) - 1; |
| 261 | } |
| 262 | |
| 263 | return -1; |
| 264 | } |
| 265 | |
| 266 | /// find_first_unset_in - Returns the index of the first unset bit in the |
| 267 | /// range [Begin, End). Returns -1 if all bits in the range are set. |
| 268 | int find_first_unset_in(unsigned Begin, unsigned End) const { |
| 269 | assert(Begin <= End && End <= Size); |
| 270 | if (Begin == End) |
| 271 | return -1; |
| 272 | |
| 273 | unsigned FirstWord = Begin / BITWORD_SIZE; |
| 274 | unsigned LastWord = (End - 1) / BITWORD_SIZE; |
| 275 | |
| 276 | // Check subsequent words. |
| 277 | for (unsigned i = FirstWord; i <= LastWord; ++i) { |
| 278 | BitWord Copy = Bits[i]; |
| 279 | |
| 280 | if (i == FirstWord) { |
| 281 | unsigned FirstBit = Begin % BITWORD_SIZE; |
| 282 | Copy |= maskTrailingOnes<BitWord>(FirstBit); |
| 283 | } |
| 284 | |
| 285 | if (i == LastWord) { |
| 286 | unsigned LastBit = (End - 1) % BITWORD_SIZE; |
| 287 | Copy |= maskTrailingZeros<BitWord>(LastBit + 1); |
| 288 | } |
| 289 | if (Copy != ~0UL) { |
| 290 | unsigned Result = i * BITWORD_SIZE + countTrailingOnes(Copy); |
| 291 | return Result < size() ? Result : -1; |
| 292 | } |
| 293 | } |
| 294 | return -1; |
| 295 | } |
| 296 | |
| 297 | /// find_last_unset_in - Returns the index of the last unset bit in the |
| 298 | /// range [Begin, End). Returns -1 if all bits in the range are set. |
| 299 | int find_last_unset_in(unsigned Begin, unsigned End) const { |
| 300 | assert(Begin <= End && End <= Size); |
| 301 | if (Begin == End) |
| 302 | return -1; |
| 303 | |
| 304 | unsigned LastWord = (End - 1) / BITWORD_SIZE; |
| 305 | unsigned FirstWord = Begin / BITWORD_SIZE; |
| 306 | |
| 307 | for (unsigned i = LastWord + 1; i >= FirstWord + 1; --i) { |
| 308 | unsigned CurrentWord = i - 1; |
| 309 | |
| 310 | BitWord Copy = Bits[CurrentWord]; |
| 311 | if (CurrentWord == LastWord) { |
| 312 | unsigned LastBit = (End - 1) % BITWORD_SIZE; |
| 313 | Copy |= maskTrailingZeros<BitWord>(LastBit + 1); |
| 314 | } |
| 315 | |
| 316 | if (CurrentWord == FirstWord) { |
| 317 | unsigned FirstBit = Begin % BITWORD_SIZE; |
| 318 | Copy |= maskTrailingOnes<BitWord>(FirstBit); |
| 319 | } |
| 320 | |
| 321 | if (Copy != ~0UL) { |
| 322 | unsigned Result = |
| 323 | (CurrentWord + 1) * BITWORD_SIZE - countLeadingOnes(Copy) - 1; |
| 324 | return Result < Size ? Result : -1; |
| 325 | } |
| 326 | } |
| 327 | return -1; |
| 328 | } |
| 329 | |
| 330 | /// find_first - Returns the index of the first set bit, -1 if none |
| 331 | /// of the bits are set. |
| 332 | int find_first() const { return find_first_in(0, Size); } |
| 333 | |
| 334 | /// find_last - Returns the index of the last set bit, -1 if none of the bits |
| 335 | /// are set. |
| 336 | int find_last() const { return find_last_in(0, Size); } |
| 337 | |
| 338 | /// find_next - Returns the index of the next set bit following the |
| 339 | /// "Prev" bit. Returns -1 if the next set bit is not found. |
| 340 | int find_next(unsigned Prev) const { return find_first_in(Prev + 1, Size); } |
| 341 | |
| 342 | /// find_prev - Returns the index of the first set bit that precedes the |
| 343 | /// the bit at \p PriorTo. Returns -1 if all previous bits are unset. |
| 344 | int find_prev(unsigned PriorTo) const { return find_last_in(0, PriorTo); } |
| 345 | |
| 346 | /// find_first_unset - Returns the index of the first unset bit, -1 if all |
| 347 | /// of the bits are set. |
| 348 | int find_first_unset() const { return find_first_unset_in(0, Size); } |
| 349 | |
| 350 | /// find_next_unset - Returns the index of the next unset bit following the |
| 351 | /// "Prev" bit. Returns -1 if all remaining bits are set. |
| 352 | int find_next_unset(unsigned Prev) const { |
| 353 | return find_first_unset_in(Prev + 1, Size); |
| 354 | } |
| 355 | |
| 356 | /// find_last_unset - Returns the index of the last unset bit, -1 if all of |
| 357 | /// the bits are set. |
| 358 | int find_last_unset() const { return find_last_unset_in(0, Size); } |
| 359 | |
| 360 | /// find_prev_unset - Returns the index of the first unset bit that precedes |
| 361 | /// the bit at \p PriorTo. Returns -1 if all previous bits are set. |
| 362 | int find_prev_unset(unsigned PriorTo) { |
| 363 | return find_last_unset_in(0, PriorTo); |
| 364 | } |
| 365 | |
| 366 | /// clear - Removes all bits from the bitvector. Does not change capacity. |
| 367 | void clear() { |
| 368 | Size = 0; |
| 369 | } |
| 370 | |
| 371 | /// resize - Grow or shrink the bitvector. |
| 372 | void resize(unsigned N, bool t = false) { |
| 373 | if (N > getBitCapacity()) { |
| 374 | unsigned OldCapacity = Bits.size(); |
| 375 | grow(N); |
| 376 | init_words(Bits.drop_front(OldCapacity), t); |
| 377 | } |
| 378 | |
| 379 | // Set any old unused bits that are now included in the BitVector. This |
| 380 | // may set bits that are not included in the new vector, but we will clear |
| 381 | // them back out below. |
| 382 | if (N > Size) |
| 383 | set_unused_bits(t); |
| 384 | |
| 385 | // Update the size, and clear out any bits that are now unused |
| 386 | unsigned OldSize = Size; |
| 387 | Size = N; |
| 388 | if (t || N < OldSize) |
| 389 | clear_unused_bits(); |
| 390 | } |
| 391 | |
| 392 | void reserve(unsigned N) { |
| 393 | if (N > getBitCapacity()) |
| 394 | grow(N); |
| 395 | } |
| 396 | |
| 397 | // Set, reset, flip |
| 398 | BitVector &set() { |
| 399 | init_words(Bits, true); |
| 400 | clear_unused_bits(); |
| 401 | return *this; |
| 402 | } |
| 403 | |
| 404 | BitVector &set(unsigned Idx) { |
| 405 | assert(Bits.data() && "Bits never allocated"); |
| 406 | Bits[Idx / BITWORD_SIZE] |= BitWord(1) << (Idx % BITWORD_SIZE); |
| 407 | return *this; |
| 408 | } |
| 409 | |
| 410 | /// set - Efficiently set a range of bits in [I, E) |
| 411 | BitVector &set(unsigned I, unsigned E) { |
| 412 | assert(I <= E && "Attempted to set backwards range!"); |
| 413 | assert(E <= size() && "Attempted to set out-of-bounds range!"); |
| 414 | |
| 415 | if (I == E) return *this; |
| 416 | |
| 417 | if (I / BITWORD_SIZE == E / BITWORD_SIZE) { |
| 418 | BitWord EMask = 1UL << (E % BITWORD_SIZE); |
| 419 | BitWord IMask = 1UL << (I % BITWORD_SIZE); |
| 420 | BitWord Mask = EMask - IMask; |
| 421 | Bits[I / BITWORD_SIZE] |= Mask; |
| 422 | return *this; |
| 423 | } |
| 424 | |
| 425 | BitWord PrefixMask = ~0UL << (I % BITWORD_SIZE); |
| 426 | Bits[I / BITWORD_SIZE] |= PrefixMask; |
| 427 | I = alignTo(I, BITWORD_SIZE); |
| 428 | |
| 429 | for (; I + BITWORD_SIZE <= E; I += BITWORD_SIZE) |
| 430 | Bits[I / BITWORD_SIZE] = ~0UL; |
| 431 | |
| 432 | BitWord PostfixMask = (1UL << (E % BITWORD_SIZE)) - 1; |
| 433 | if (I < E) |
| 434 | Bits[I / BITWORD_SIZE] |= PostfixMask; |
| 435 | |
| 436 | return *this; |
| 437 | } |
| 438 | |
| 439 | BitVector &reset() { |
| 440 | init_words(Bits, false); |
| 441 | return *this; |
| 442 | } |
| 443 | |
| 444 | BitVector &reset(unsigned Idx) { |
| 445 | Bits[Idx / BITWORD_SIZE] &= ~(BitWord(1) << (Idx % BITWORD_SIZE)); |
| 446 | return *this; |
| 447 | } |
| 448 | |
| 449 | /// reset - Efficiently reset a range of bits in [I, E) |
| 450 | BitVector &reset(unsigned I, unsigned E) { |
| 451 | assert(I <= E && "Attempted to reset backwards range!"); |
| 452 | assert(E <= size() && "Attempted to reset out-of-bounds range!"); |
| 453 | |
| 454 | if (I == E) return *this; |
| 455 | |
| 456 | if (I / BITWORD_SIZE == E / BITWORD_SIZE) { |
| 457 | BitWord EMask = 1UL << (E % BITWORD_SIZE); |
| 458 | BitWord IMask = 1UL << (I % BITWORD_SIZE); |
| 459 | BitWord Mask = EMask - IMask; |
| 460 | Bits[I / BITWORD_SIZE] &= ~Mask; |
| 461 | return *this; |
| 462 | } |
| 463 | |
| 464 | BitWord PrefixMask = ~0UL << (I % BITWORD_SIZE); |
| 465 | Bits[I / BITWORD_SIZE] &= ~PrefixMask; |
| 466 | I = alignTo(I, BITWORD_SIZE); |
| 467 | |
| 468 | for (; I + BITWORD_SIZE <= E; I += BITWORD_SIZE) |
| 469 | Bits[I / BITWORD_SIZE] = 0UL; |
| 470 | |
| 471 | BitWord PostfixMask = (1UL << (E % BITWORD_SIZE)) - 1; |
| 472 | if (I < E) |
| 473 | Bits[I / BITWORD_SIZE] &= ~PostfixMask; |
| 474 | |
| 475 | return *this; |
| 476 | } |
| 477 | |
| 478 | BitVector &flip() { |
| 479 | for (unsigned i = 0; i < NumBitWords(size()); ++i) |
| 480 | Bits[i] = ~Bits[i]; |
| 481 | clear_unused_bits(); |
| 482 | return *this; |
| 483 | } |
| 484 | |
| 485 | BitVector &flip(unsigned Idx) { |
| 486 | Bits[Idx / BITWORD_SIZE] ^= BitWord(1) << (Idx % BITWORD_SIZE); |
| 487 | return *this; |
| 488 | } |
| 489 | |
| 490 | // Indexing. |
| 491 | reference operator[](unsigned Idx) { |
| 492 | assert (Idx < Size && "Out-of-bounds Bit access."); |
| 493 | return reference(*this, Idx); |
| 494 | } |
| 495 | |
| 496 | bool operator[](unsigned Idx) const { |
| 497 | assert (Idx < Size && "Out-of-bounds Bit access."); |
| 498 | BitWord Mask = BitWord(1) << (Idx % BITWORD_SIZE); |
| 499 | return (Bits[Idx / BITWORD_SIZE] & Mask) != 0; |
| 500 | } |
| 501 | |
| 502 | bool test(unsigned Idx) const { |
| 503 | return (*this)[Idx]; |
| 504 | } |
| 505 | |
Andrew Scull | 0372a57 | 2018-11-16 15:47:06 +0000 | [diff] [blame] | 506 | // Push single bit to end of vector. |
| 507 | void push_back(bool Val) { |
| 508 | unsigned OldSize = Size; |
| 509 | unsigned NewSize = Size + 1; |
| 510 | |
| 511 | // Resize, which will insert zeros. |
| 512 | // If we already fit then the unused bits will be already zero. |
| 513 | if (NewSize > getBitCapacity()) |
| 514 | resize(NewSize, false); |
| 515 | else |
| 516 | Size = NewSize; |
| 517 | |
| 518 | // If true, set single bit. |
| 519 | if (Val) |
| 520 | set(OldSize); |
| 521 | } |
| 522 | |
Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 523 | /// Test if any common bits are set. |
| 524 | bool anyCommon(const BitVector &RHS) const { |
| 525 | unsigned ThisWords = NumBitWords(size()); |
| 526 | unsigned RHSWords = NumBitWords(RHS.size()); |
| 527 | for (unsigned i = 0, e = std::min(ThisWords, RHSWords); i != e; ++i) |
| 528 | if (Bits[i] & RHS.Bits[i]) |
| 529 | return true; |
| 530 | return false; |
| 531 | } |
| 532 | |
| 533 | // Comparison operators. |
| 534 | bool operator==(const BitVector &RHS) const { |
| 535 | unsigned ThisWords = NumBitWords(size()); |
| 536 | unsigned RHSWords = NumBitWords(RHS.size()); |
| 537 | unsigned i; |
| 538 | for (i = 0; i != std::min(ThisWords, RHSWords); ++i) |
| 539 | if (Bits[i] != RHS.Bits[i]) |
| 540 | return false; |
| 541 | |
| 542 | // Verify that any extra words are all zeros. |
| 543 | if (i != ThisWords) { |
| 544 | for (; i != ThisWords; ++i) |
| 545 | if (Bits[i]) |
| 546 | return false; |
| 547 | } else if (i != RHSWords) { |
| 548 | for (; i != RHSWords; ++i) |
| 549 | if (RHS.Bits[i]) |
| 550 | return false; |
| 551 | } |
| 552 | return true; |
| 553 | } |
| 554 | |
| 555 | bool operator!=(const BitVector &RHS) const { |
| 556 | return !(*this == RHS); |
| 557 | } |
| 558 | |
| 559 | /// Intersection, union, disjoint union. |
| 560 | BitVector &operator&=(const BitVector &RHS) { |
| 561 | unsigned ThisWords = NumBitWords(size()); |
| 562 | unsigned RHSWords = NumBitWords(RHS.size()); |
| 563 | unsigned i; |
| 564 | for (i = 0; i != std::min(ThisWords, RHSWords); ++i) |
| 565 | Bits[i] &= RHS.Bits[i]; |
| 566 | |
| 567 | // Any bits that are just in this bitvector become zero, because they aren't |
| 568 | // in the RHS bit vector. Any words only in RHS are ignored because they |
| 569 | // are already zero in the LHS. |
| 570 | for (; i != ThisWords; ++i) |
| 571 | Bits[i] = 0; |
| 572 | |
| 573 | return *this; |
| 574 | } |
| 575 | |
| 576 | /// reset - Reset bits that are set in RHS. Same as *this &= ~RHS. |
| 577 | BitVector &reset(const BitVector &RHS) { |
| 578 | unsigned ThisWords = NumBitWords(size()); |
| 579 | unsigned RHSWords = NumBitWords(RHS.size()); |
| 580 | unsigned i; |
| 581 | for (i = 0; i != std::min(ThisWords, RHSWords); ++i) |
| 582 | Bits[i] &= ~RHS.Bits[i]; |
| 583 | return *this; |
| 584 | } |
| 585 | |
| 586 | /// test - Check if (This - RHS) is zero. |
| 587 | /// This is the same as reset(RHS) and any(). |
| 588 | bool test(const BitVector &RHS) const { |
| 589 | unsigned ThisWords = NumBitWords(size()); |
| 590 | unsigned RHSWords = NumBitWords(RHS.size()); |
| 591 | unsigned i; |
| 592 | for (i = 0; i != std::min(ThisWords, RHSWords); ++i) |
| 593 | if ((Bits[i] & ~RHS.Bits[i]) != 0) |
| 594 | return true; |
| 595 | |
| 596 | for (; i != ThisWords ; ++i) |
| 597 | if (Bits[i] != 0) |
| 598 | return true; |
| 599 | |
| 600 | return false; |
| 601 | } |
| 602 | |
| 603 | BitVector &operator|=(const BitVector &RHS) { |
| 604 | if (size() < RHS.size()) |
| 605 | resize(RHS.size()); |
| 606 | for (size_t i = 0, e = NumBitWords(RHS.size()); i != e; ++i) |
| 607 | Bits[i] |= RHS.Bits[i]; |
| 608 | return *this; |
| 609 | } |
| 610 | |
| 611 | BitVector &operator^=(const BitVector &RHS) { |
| 612 | if (size() < RHS.size()) |
| 613 | resize(RHS.size()); |
| 614 | for (size_t i = 0, e = NumBitWords(RHS.size()); i != e; ++i) |
| 615 | Bits[i] ^= RHS.Bits[i]; |
| 616 | return *this; |
| 617 | } |
| 618 | |
| 619 | BitVector &operator>>=(unsigned N) { |
| 620 | assert(N <= Size); |
| 621 | if (LLVM_UNLIKELY(empty() || N == 0)) |
| 622 | return *this; |
| 623 | |
| 624 | unsigned NumWords = NumBitWords(Size); |
| 625 | assert(NumWords >= 1); |
| 626 | |
| 627 | wordShr(N / BITWORD_SIZE); |
| 628 | |
| 629 | unsigned BitDistance = N % BITWORD_SIZE; |
| 630 | if (BitDistance == 0) |
| 631 | return *this; |
| 632 | |
| 633 | // When the shift size is not a multiple of the word size, then we have |
| 634 | // a tricky situation where each word in succession needs to extract some |
| 635 | // of the bits from the next word and or them into this word while |
| 636 | // shifting this word to make room for the new bits. This has to be done |
| 637 | // for every word in the array. |
| 638 | |
| 639 | // Since we're shifting each word right, some bits will fall off the end |
| 640 | // of each word to the right, and empty space will be created on the left. |
| 641 | // The final word in the array will lose bits permanently, so starting at |
| 642 | // the beginning, work forwards shifting each word to the right, and |
| 643 | // OR'ing in the bits from the end of the next word to the beginning of |
| 644 | // the current word. |
| 645 | |
| 646 | // Example: |
| 647 | // Starting with {0xAABBCCDD, 0xEEFF0011, 0x22334455} and shifting right |
| 648 | // by 4 bits. |
| 649 | // Step 1: Word[0] >>= 4 ; 0x0ABBCCDD |
| 650 | // Step 2: Word[0] |= 0x10000000 ; 0x1ABBCCDD |
| 651 | // Step 3: Word[1] >>= 4 ; 0x0EEFF001 |
| 652 | // Step 4: Word[1] |= 0x50000000 ; 0x5EEFF001 |
| 653 | // Step 5: Word[2] >>= 4 ; 0x02334455 |
| 654 | // Result: { 0x1ABBCCDD, 0x5EEFF001, 0x02334455 } |
| 655 | const BitWord Mask = maskTrailingOnes<BitWord>(BitDistance); |
| 656 | const unsigned LSH = BITWORD_SIZE - BitDistance; |
| 657 | |
| 658 | for (unsigned I = 0; I < NumWords - 1; ++I) { |
| 659 | Bits[I] >>= BitDistance; |
| 660 | Bits[I] |= (Bits[I + 1] & Mask) << LSH; |
| 661 | } |
| 662 | |
| 663 | Bits[NumWords - 1] >>= BitDistance; |
| 664 | |
| 665 | return *this; |
| 666 | } |
| 667 | |
| 668 | BitVector &operator<<=(unsigned N) { |
| 669 | assert(N <= Size); |
| 670 | if (LLVM_UNLIKELY(empty() || N == 0)) |
| 671 | return *this; |
| 672 | |
| 673 | unsigned NumWords = NumBitWords(Size); |
| 674 | assert(NumWords >= 1); |
| 675 | |
| 676 | wordShl(N / BITWORD_SIZE); |
| 677 | |
| 678 | unsigned BitDistance = N % BITWORD_SIZE; |
| 679 | if (BitDistance == 0) |
| 680 | return *this; |
| 681 | |
| 682 | // When the shift size is not a multiple of the word size, then we have |
| 683 | // a tricky situation where each word in succession needs to extract some |
| 684 | // of the bits from the previous word and or them into this word while |
| 685 | // shifting this word to make room for the new bits. This has to be done |
| 686 | // for every word in the array. This is similar to the algorithm outlined |
| 687 | // in operator>>=, but backwards. |
| 688 | |
| 689 | // Since we're shifting each word left, some bits will fall off the end |
| 690 | // of each word to the left, and empty space will be created on the right. |
| 691 | // The first word in the array will lose bits permanently, so starting at |
| 692 | // the end, work backwards shifting each word to the left, and OR'ing |
| 693 | // in the bits from the end of the next word to the beginning of the |
| 694 | // current word. |
| 695 | |
| 696 | // Example: |
| 697 | // Starting with {0xAABBCCDD, 0xEEFF0011, 0x22334455} and shifting left |
| 698 | // by 4 bits. |
| 699 | // Step 1: Word[2] <<= 4 ; 0x23344550 |
| 700 | // Step 2: Word[2] |= 0x0000000E ; 0x2334455E |
| 701 | // Step 3: Word[1] <<= 4 ; 0xEFF00110 |
| 702 | // Step 4: Word[1] |= 0x0000000A ; 0xEFF0011A |
| 703 | // Step 5: Word[0] <<= 4 ; 0xABBCCDD0 |
| 704 | // Result: { 0xABBCCDD0, 0xEFF0011A, 0x2334455E } |
| 705 | const BitWord Mask = maskLeadingOnes<BitWord>(BitDistance); |
| 706 | const unsigned RSH = BITWORD_SIZE - BitDistance; |
| 707 | |
| 708 | for (int I = NumWords - 1; I > 0; --I) { |
| 709 | Bits[I] <<= BitDistance; |
| 710 | Bits[I] |= (Bits[I - 1] & Mask) >> RSH; |
| 711 | } |
| 712 | Bits[0] <<= BitDistance; |
| 713 | clear_unused_bits(); |
| 714 | |
| 715 | return *this; |
| 716 | } |
| 717 | |
| 718 | // Assignment operator. |
| 719 | const BitVector &operator=(const BitVector &RHS) { |
| 720 | if (this == &RHS) return *this; |
| 721 | |
| 722 | Size = RHS.size(); |
| 723 | unsigned RHSWords = NumBitWords(Size); |
| 724 | if (Size <= getBitCapacity()) { |
| 725 | if (Size) |
| 726 | std::memcpy(Bits.data(), RHS.Bits.data(), RHSWords * sizeof(BitWord)); |
| 727 | clear_unused_bits(); |
| 728 | return *this; |
| 729 | } |
| 730 | |
| 731 | // Grow the bitvector to have enough elements. |
| 732 | unsigned NewCapacity = RHSWords; |
| 733 | assert(NewCapacity > 0 && "negative capacity?"); |
| 734 | auto NewBits = allocate(NewCapacity); |
| 735 | std::memcpy(NewBits.data(), RHS.Bits.data(), NewCapacity * sizeof(BitWord)); |
| 736 | |
| 737 | // Destroy the old bits. |
| 738 | std::free(Bits.data()); |
| 739 | Bits = NewBits; |
| 740 | |
| 741 | return *this; |
| 742 | } |
| 743 | |
| 744 | const BitVector &operator=(BitVector &&RHS) { |
| 745 | if (this == &RHS) return *this; |
| 746 | |
| 747 | std::free(Bits.data()); |
| 748 | Bits = RHS.Bits; |
| 749 | Size = RHS.Size; |
| 750 | |
| 751 | RHS.Bits = MutableArrayRef<BitWord>(); |
| 752 | RHS.Size = 0; |
| 753 | |
| 754 | return *this; |
| 755 | } |
| 756 | |
| 757 | void swap(BitVector &RHS) { |
| 758 | std::swap(Bits, RHS.Bits); |
| 759 | std::swap(Size, RHS.Size); |
| 760 | } |
| 761 | |
| 762 | //===--------------------------------------------------------------------===// |
| 763 | // Portable bit mask operations. |
| 764 | //===--------------------------------------------------------------------===// |
| 765 | // |
| 766 | // These methods all operate on arrays of uint32_t, each holding 32 bits. The |
| 767 | // fixed word size makes it easier to work with literal bit vector constants |
| 768 | // in portable code. |
| 769 | // |
| 770 | // The LSB in each word is the lowest numbered bit. The size of a portable |
| 771 | // bit mask is always a whole multiple of 32 bits. If no bit mask size is |
| 772 | // given, the bit mask is assumed to cover the entire BitVector. |
| 773 | |
| 774 | /// setBitsInMask - Add '1' bits from Mask to this vector. Don't resize. |
| 775 | /// This computes "*this |= Mask". |
| 776 | void setBitsInMask(const uint32_t *Mask, unsigned MaskWords = ~0u) { |
| 777 | applyMask<true, false>(Mask, MaskWords); |
| 778 | } |
| 779 | |
| 780 | /// clearBitsInMask - Clear any bits in this vector that are set in Mask. |
| 781 | /// Don't resize. This computes "*this &= ~Mask". |
| 782 | void clearBitsInMask(const uint32_t *Mask, unsigned MaskWords = ~0u) { |
| 783 | applyMask<false, false>(Mask, MaskWords); |
| 784 | } |
| 785 | |
| 786 | /// setBitsNotInMask - Add a bit to this vector for every '0' bit in Mask. |
| 787 | /// Don't resize. This computes "*this |= ~Mask". |
| 788 | void setBitsNotInMask(const uint32_t *Mask, unsigned MaskWords = ~0u) { |
| 789 | applyMask<true, true>(Mask, MaskWords); |
| 790 | } |
| 791 | |
| 792 | /// clearBitsNotInMask - Clear a bit in this vector for every '0' bit in Mask. |
| 793 | /// Don't resize. This computes "*this &= Mask". |
| 794 | void clearBitsNotInMask(const uint32_t *Mask, unsigned MaskWords = ~0u) { |
| 795 | applyMask<false, true>(Mask, MaskWords); |
| 796 | } |
| 797 | |
| 798 | private: |
Andrew Scull | cdfcccc | 2018-10-05 20:58:37 +0100 | [diff] [blame] | 799 | /// Perform a logical left shift of \p Count words by moving everything |
Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 800 | /// \p Count words to the right in memory. |
| 801 | /// |
| 802 | /// While confusing, words are stored from least significant at Bits[0] to |
| 803 | /// most significant at Bits[NumWords-1]. A logical shift left, however, |
| 804 | /// moves the current least significant bit to a higher logical index, and |
| 805 | /// fills the previous least significant bits with 0. Thus, we actually |
| 806 | /// need to move the bytes of the memory to the right, not to the left. |
| 807 | /// Example: |
| 808 | /// Words = [0xBBBBAAAA, 0xDDDDFFFF, 0x00000000, 0xDDDD0000] |
| 809 | /// represents a BitVector where 0xBBBBAAAA contain the least significant |
| 810 | /// bits. So if we want to shift the BitVector left by 2 words, we need to |
| 811 | /// turn this into 0x00000000 0x00000000 0xBBBBAAAA 0xDDDDFFFF by using a |
| 812 | /// memmove which moves right, not left. |
| 813 | void wordShl(uint32_t Count) { |
| 814 | if (Count == 0) |
| 815 | return; |
| 816 | |
| 817 | uint32_t NumWords = NumBitWords(Size); |
| 818 | |
| 819 | auto Src = Bits.take_front(NumWords).drop_back(Count); |
| 820 | auto Dest = Bits.take_front(NumWords).drop_front(Count); |
| 821 | |
| 822 | // Since we always move Word-sized chunks of data with src and dest both |
| 823 | // aligned to a word-boundary, we don't need to worry about endianness |
| 824 | // here. |
| 825 | std::memmove(Dest.begin(), Src.begin(), Dest.size() * sizeof(BitWord)); |
| 826 | std::memset(Bits.data(), 0, Count * sizeof(BitWord)); |
| 827 | clear_unused_bits(); |
| 828 | } |
| 829 | |
Andrew Scull | cdfcccc | 2018-10-05 20:58:37 +0100 | [diff] [blame] | 830 | /// Perform a logical right shift of \p Count words by moving those |
Andrew Scull | 5e1ddfa | 2018-08-14 10:06:54 +0100 | [diff] [blame] | 831 | /// words to the left in memory. See wordShl for more information. |
| 832 | /// |
| 833 | void wordShr(uint32_t Count) { |
| 834 | if (Count == 0) |
| 835 | return; |
| 836 | |
| 837 | uint32_t NumWords = NumBitWords(Size); |
| 838 | |
| 839 | auto Src = Bits.take_front(NumWords).drop_front(Count); |
| 840 | auto Dest = Bits.take_front(NumWords).drop_back(Count); |
| 841 | assert(Dest.size() == Src.size()); |
| 842 | |
| 843 | std::memmove(Dest.begin(), Src.begin(), Dest.size() * sizeof(BitWord)); |
| 844 | std::memset(Dest.end(), 0, Count * sizeof(BitWord)); |
| 845 | } |
| 846 | |
| 847 | MutableArrayRef<BitWord> allocate(size_t NumWords) { |
| 848 | BitWord *RawBits = static_cast<BitWord *>( |
| 849 | safe_malloc(NumWords * sizeof(BitWord))); |
| 850 | return MutableArrayRef<BitWord>(RawBits, NumWords); |
| 851 | } |
| 852 | |
| 853 | int next_unset_in_word(int WordIndex, BitWord Word) const { |
| 854 | unsigned Result = WordIndex * BITWORD_SIZE + countTrailingOnes(Word); |
| 855 | return Result < size() ? Result : -1; |
| 856 | } |
| 857 | |
| 858 | unsigned NumBitWords(unsigned S) const { |
| 859 | return (S + BITWORD_SIZE-1) / BITWORD_SIZE; |
| 860 | } |
| 861 | |
| 862 | // Set the unused bits in the high words. |
| 863 | void set_unused_bits(bool t = true) { |
| 864 | // Set high words first. |
| 865 | unsigned UsedWords = NumBitWords(Size); |
| 866 | if (Bits.size() > UsedWords) |
| 867 | init_words(Bits.drop_front(UsedWords), t); |
| 868 | |
| 869 | // Then set any stray high bits of the last used word. |
| 870 | unsigned ExtraBits = Size % BITWORD_SIZE; |
| 871 | if (ExtraBits) { |
| 872 | BitWord ExtraBitMask = ~0UL << ExtraBits; |
| 873 | if (t) |
| 874 | Bits[UsedWords-1] |= ExtraBitMask; |
| 875 | else |
| 876 | Bits[UsedWords-1] &= ~ExtraBitMask; |
| 877 | } |
| 878 | } |
| 879 | |
| 880 | // Clear the unused bits in the high words. |
| 881 | void clear_unused_bits() { |
| 882 | set_unused_bits(false); |
| 883 | } |
| 884 | |
| 885 | void grow(unsigned NewSize) { |
| 886 | size_t NewCapacity = std::max<size_t>(NumBitWords(NewSize), Bits.size() * 2); |
| 887 | assert(NewCapacity > 0 && "realloc-ing zero space"); |
| 888 | BitWord *NewBits = static_cast<BitWord *>( |
| 889 | safe_realloc(Bits.data(), NewCapacity * sizeof(BitWord))); |
| 890 | Bits = MutableArrayRef<BitWord>(NewBits, NewCapacity); |
| 891 | clear_unused_bits(); |
| 892 | } |
| 893 | |
| 894 | void init_words(MutableArrayRef<BitWord> B, bool t) { |
| 895 | if (B.size() > 0) |
| 896 | memset(B.data(), 0 - (int)t, B.size() * sizeof(BitWord)); |
| 897 | } |
| 898 | |
| 899 | template<bool AddBits, bool InvertMask> |
| 900 | void applyMask(const uint32_t *Mask, unsigned MaskWords) { |
| 901 | static_assert(BITWORD_SIZE % 32 == 0, "Unsupported BitWord size."); |
| 902 | MaskWords = std::min(MaskWords, (size() + 31) / 32); |
| 903 | const unsigned Scale = BITWORD_SIZE / 32; |
| 904 | unsigned i; |
| 905 | for (i = 0; MaskWords >= Scale; ++i, MaskWords -= Scale) { |
| 906 | BitWord BW = Bits[i]; |
| 907 | // This inner loop should unroll completely when BITWORD_SIZE > 32. |
| 908 | for (unsigned b = 0; b != BITWORD_SIZE; b += 32) { |
| 909 | uint32_t M = *Mask++; |
| 910 | if (InvertMask) M = ~M; |
| 911 | if (AddBits) BW |= BitWord(M) << b; |
| 912 | else BW &= ~(BitWord(M) << b); |
| 913 | } |
| 914 | Bits[i] = BW; |
| 915 | } |
| 916 | for (unsigned b = 0; MaskWords; b += 32, --MaskWords) { |
| 917 | uint32_t M = *Mask++; |
| 918 | if (InvertMask) M = ~M; |
| 919 | if (AddBits) Bits[i] |= BitWord(M) << b; |
| 920 | else Bits[i] &= ~(BitWord(M) << b); |
| 921 | } |
| 922 | if (AddBits) |
| 923 | clear_unused_bits(); |
| 924 | } |
| 925 | |
| 926 | public: |
| 927 | /// Return the size (in bytes) of the bit vector. |
| 928 | size_t getMemorySize() const { return Bits.size() * sizeof(BitWord); } |
| 929 | size_t getBitCapacity() const { return Bits.size() * BITWORD_SIZE; } |
| 930 | }; |
| 931 | |
| 932 | inline size_t capacity_in_bytes(const BitVector &X) { |
| 933 | return X.getMemorySize(); |
| 934 | } |
| 935 | |
| 936 | } // end namespace llvm |
| 937 | |
| 938 | namespace std { |
| 939 | /// Implement std::swap in terms of BitVector swap. |
| 940 | inline void |
| 941 | swap(llvm::BitVector &LHS, llvm::BitVector &RHS) { |
| 942 | LHS.swap(RHS); |
| 943 | } |
| 944 | } // end namespace std |
| 945 | |
| 946 | #endif // LLVM_ADT_BITVECTOR_H |