BitVector.h revision dce4a407a24b04eebc6a376f8e62b41aaa7b071f
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/Support/Compiler.h" 18#include "llvm/Support/ErrorHandling.h" 19#include "llvm/Support/MathExtras.h" 20#include <algorithm> 21#include <cassert> 22#include <climits> 23#include <cstdlib> 24 25namespace llvm { 26 27class BitVector { 28 typedef unsigned long BitWord; 29 30 enum { BITWORD_SIZE = (unsigned)sizeof(BitWord) * CHAR_BIT }; 31 32 BitWord *Bits; // Actual bits. 33 unsigned Size; // Size of bitvector in bits. 34 unsigned Capacity; // Size of allocated memory in BitWord. 35 36public: 37 // Encapsulation of a single bit. 38 class reference { 39 friend class BitVector; 40 41 BitWord *WordRef; 42 unsigned BitPos; 43 44 reference(); // Undefined 45 46 public: 47 reference(BitVector &b, unsigned Idx) { 48 WordRef = &b.Bits[Idx / BITWORD_SIZE]; 49 BitPos = Idx % BITWORD_SIZE; 50 } 51 52 ~reference() {} 53 54 reference &operator=(reference t) { 55 *this = bool(t); 56 return *this; 57 } 58 59 reference& operator=(bool t) { 60 if (t) 61 *WordRef |= BitWord(1) << BitPos; 62 else 63 *WordRef &= ~(BitWord(1) << BitPos); 64 return *this; 65 } 66 67 operator bool() const { 68 return ((*WordRef) & (BitWord(1) << BitPos)) ? true : false; 69 } 70 }; 71 72 73 /// BitVector default ctor - Creates an empty bitvector. 74 BitVector() : Size(0), Capacity(0) { 75 Bits = nullptr; 76 } 77 78 /// BitVector ctor - Creates a bitvector of specified number of bits. All 79 /// bits are initialized to the specified value. 80 explicit BitVector(unsigned s, bool t = false) : Size(s) { 81 Capacity = NumBitWords(s); 82 Bits = (BitWord *)std::malloc(Capacity * sizeof(BitWord)); 83 init_words(Bits, Capacity, t); 84 if (t) 85 clear_unused_bits(); 86 } 87 88 /// BitVector copy ctor. 89 BitVector(const BitVector &RHS) : Size(RHS.size()) { 90 if (Size == 0) { 91 Bits = nullptr; 92 Capacity = 0; 93 return; 94 } 95 96 Capacity = NumBitWords(RHS.size()); 97 Bits = (BitWord *)std::malloc(Capacity * sizeof(BitWord)); 98 std::memcpy(Bits, RHS.Bits, Capacity * sizeof(BitWord)); 99 } 100 101 BitVector(BitVector &&RHS) 102 : Bits(RHS.Bits), Size(RHS.Size), Capacity(RHS.Capacity) { 103 RHS.Bits = nullptr; 104 } 105 106 ~BitVector() { 107 std::free(Bits); 108 } 109 110 /// empty - Tests whether there are no bits in this bitvector. 111 bool empty() const { return Size == 0; } 112 113 /// size - Returns the number of bits in this bitvector. 114 unsigned size() const { return Size; } 115 116 /// count - Returns the number of bits which are set. 117 unsigned count() const { 118 unsigned NumBits = 0; 119 for (unsigned i = 0; i < NumBitWords(size()); ++i) 120 if (sizeof(BitWord) == 4) 121 NumBits += CountPopulation_32((uint32_t)Bits[i]); 122 else if (sizeof(BitWord) == 8) 123 NumBits += CountPopulation_64(Bits[i]); 124 else 125 llvm_unreachable("Unsupported!"); 126 return NumBits; 127 } 128 129 /// any - Returns true if any bit is set. 130 bool any() const { 131 for (unsigned i = 0; i < NumBitWords(size()); ++i) 132 if (Bits[i] != 0) 133 return true; 134 return false; 135 } 136 137 /// all - Returns true if all bits are set. 138 bool all() const { 139 for (unsigned i = 0; i < Size / BITWORD_SIZE; ++i) 140 if (Bits[i] != ~0UL) 141 return false; 142 143 // If bits remain check that they are ones. The unused bits are always zero. 144 if (unsigned Remainder = Size % BITWORD_SIZE) 145 return Bits[Size / BITWORD_SIZE] == (1UL << Remainder) - 1; 146 147 return true; 148 } 149 150 /// none - Returns true if none of the bits are set. 151 bool none() const { 152 return !any(); 153 } 154 155 /// find_first - Returns the index of the first set bit, -1 if none 156 /// of the bits are set. 157 int find_first() const { 158 for (unsigned i = 0; i < NumBitWords(size()); ++i) 159 if (Bits[i] != 0) { 160 if (sizeof(BitWord) == 4) 161 return i * BITWORD_SIZE + countTrailingZeros((uint32_t)Bits[i]); 162 if (sizeof(BitWord) == 8) 163 return i * BITWORD_SIZE + countTrailingZeros(Bits[i]); 164 llvm_unreachable("Unsupported!"); 165 } 166 return -1; 167 } 168 169 /// find_next - Returns the index of the next set bit following the 170 /// "Prev" bit. Returns -1 if the next set bit is not found. 171 int find_next(unsigned Prev) const { 172 ++Prev; 173 if (Prev >= Size) 174 return -1; 175 176 unsigned WordPos = Prev / BITWORD_SIZE; 177 unsigned BitPos = Prev % BITWORD_SIZE; 178 BitWord Copy = Bits[WordPos]; 179 // Mask off previous bits. 180 Copy &= ~0UL << BitPos; 181 182 if (Copy != 0) { 183 if (sizeof(BitWord) == 4) 184 return WordPos * BITWORD_SIZE + countTrailingZeros((uint32_t)Copy); 185 if (sizeof(BitWord) == 8) 186 return WordPos * BITWORD_SIZE + countTrailingZeros(Copy); 187 llvm_unreachable("Unsupported!"); 188 } 189 190 // Check subsequent words. 191 for (unsigned i = WordPos+1; i < NumBitWords(size()); ++i) 192 if (Bits[i] != 0) { 193 if (sizeof(BitWord) == 4) 194 return i * BITWORD_SIZE + countTrailingZeros((uint32_t)Bits[i]); 195 if (sizeof(BitWord) == 8) 196 return i * BITWORD_SIZE + countTrailingZeros(Bits[i]); 197 llvm_unreachable("Unsupported!"); 198 } 199 return -1; 200 } 201 202 /// clear - Clear all bits. 203 void clear() { 204 Size = 0; 205 } 206 207 /// resize - Grow or shrink the bitvector. 208 void resize(unsigned N, bool t = false) { 209 if (N > Capacity * BITWORD_SIZE) { 210 unsigned OldCapacity = Capacity; 211 grow(N); 212 init_words(&Bits[OldCapacity], (Capacity-OldCapacity), t); 213 } 214 215 // Set any old unused bits that are now included in the BitVector. This 216 // may set bits that are not included in the new vector, but we will clear 217 // them back out below. 218 if (N > Size) 219 set_unused_bits(t); 220 221 // Update the size, and clear out any bits that are now unused 222 unsigned OldSize = Size; 223 Size = N; 224 if (t || N < OldSize) 225 clear_unused_bits(); 226 } 227 228 void reserve(unsigned N) { 229 if (N > Capacity * BITWORD_SIZE) 230 grow(N); 231 } 232 233 // Set, reset, flip 234 BitVector &set() { 235 init_words(Bits, Capacity, true); 236 clear_unused_bits(); 237 return *this; 238 } 239 240 BitVector &set(unsigned Idx) { 241 Bits[Idx / BITWORD_SIZE] |= BitWord(1) << (Idx % BITWORD_SIZE); 242 return *this; 243 } 244 245 /// set - Efficiently set a range of bits in [I, E) 246 BitVector &set(unsigned I, unsigned E) { 247 assert(I <= E && "Attempted to set backwards range!"); 248 assert(E <= size() && "Attempted to set out-of-bounds range!"); 249 250 if (I == E) return *this; 251 252 if (I / BITWORD_SIZE == E / BITWORD_SIZE) { 253 BitWord EMask = 1UL << (E % BITWORD_SIZE); 254 BitWord IMask = 1UL << (I % BITWORD_SIZE); 255 BitWord Mask = EMask - IMask; 256 Bits[I / BITWORD_SIZE] |= Mask; 257 return *this; 258 } 259 260 BitWord PrefixMask = ~0UL << (I % BITWORD_SIZE); 261 Bits[I / BITWORD_SIZE] |= PrefixMask; 262 I = RoundUpToAlignment(I, BITWORD_SIZE); 263 264 for (; I + BITWORD_SIZE <= E; I += BITWORD_SIZE) 265 Bits[I / BITWORD_SIZE] = ~0UL; 266 267 BitWord PostfixMask = (1UL << (E % BITWORD_SIZE)) - 1; 268 if (I < E) 269 Bits[I / BITWORD_SIZE] |= PostfixMask; 270 271 return *this; 272 } 273 274 BitVector &reset() { 275 init_words(Bits, Capacity, false); 276 return *this; 277 } 278 279 BitVector &reset(unsigned Idx) { 280 Bits[Idx / BITWORD_SIZE] &= ~(BitWord(1) << (Idx % BITWORD_SIZE)); 281 return *this; 282 } 283 284 /// reset - Efficiently reset a range of bits in [I, E) 285 BitVector &reset(unsigned I, unsigned E) { 286 assert(I <= E && "Attempted to reset backwards range!"); 287 assert(E <= size() && "Attempted to reset out-of-bounds range!"); 288 289 if (I == E) return *this; 290 291 if (I / BITWORD_SIZE == E / BITWORD_SIZE) { 292 BitWord EMask = 1UL << (E % BITWORD_SIZE); 293 BitWord IMask = 1UL << (I % BITWORD_SIZE); 294 BitWord Mask = EMask - IMask; 295 Bits[I / BITWORD_SIZE] &= ~Mask; 296 return *this; 297 } 298 299 BitWord PrefixMask = ~0UL << (I % BITWORD_SIZE); 300 Bits[I / BITWORD_SIZE] &= ~PrefixMask; 301 I = RoundUpToAlignment(I, BITWORD_SIZE); 302 303 for (; I + BITWORD_SIZE <= E; I += BITWORD_SIZE) 304 Bits[I / BITWORD_SIZE] = 0UL; 305 306 BitWord PostfixMask = (1UL << (E % BITWORD_SIZE)) - 1; 307 if (I < E) 308 Bits[I / BITWORD_SIZE] &= ~PostfixMask; 309 310 return *this; 311 } 312 313 BitVector &flip() { 314 for (unsigned i = 0; i < NumBitWords(size()); ++i) 315 Bits[i] = ~Bits[i]; 316 clear_unused_bits(); 317 return *this; 318 } 319 320 BitVector &flip(unsigned Idx) { 321 Bits[Idx / BITWORD_SIZE] ^= BitWord(1) << (Idx % BITWORD_SIZE); 322 return *this; 323 } 324 325 // Indexing. 326 reference operator[](unsigned Idx) { 327 assert (Idx < Size && "Out-of-bounds Bit access."); 328 return reference(*this, Idx); 329 } 330 331 bool operator[](unsigned Idx) const { 332 assert (Idx < Size && "Out-of-bounds Bit access."); 333 BitWord Mask = BitWord(1) << (Idx % BITWORD_SIZE); 334 return (Bits[Idx / BITWORD_SIZE] & Mask) != 0; 335 } 336 337 bool test(unsigned Idx) const { 338 return (*this)[Idx]; 339 } 340 341 /// Test if any common bits are set. 342 bool anyCommon(const BitVector &RHS) const { 343 unsigned ThisWords = NumBitWords(size()); 344 unsigned RHSWords = NumBitWords(RHS.size()); 345 for (unsigned i = 0, e = std::min(ThisWords, RHSWords); i != e; ++i) 346 if (Bits[i] & RHS.Bits[i]) 347 return true; 348 return false; 349 } 350 351 // Comparison operators. 352 bool operator==(const BitVector &RHS) const { 353 unsigned ThisWords = NumBitWords(size()); 354 unsigned RHSWords = NumBitWords(RHS.size()); 355 unsigned i; 356 for (i = 0; i != std::min(ThisWords, RHSWords); ++i) 357 if (Bits[i] != RHS.Bits[i]) 358 return false; 359 360 // Verify that any extra words are all zeros. 361 if (i != ThisWords) { 362 for (; i != ThisWords; ++i) 363 if (Bits[i]) 364 return false; 365 } else if (i != RHSWords) { 366 for (; i != RHSWords; ++i) 367 if (RHS.Bits[i]) 368 return false; 369 } 370 return true; 371 } 372 373 bool operator!=(const BitVector &RHS) const { 374 return !(*this == RHS); 375 } 376 377 /// Intersection, union, disjoint union. 378 BitVector &operator&=(const BitVector &RHS) { 379 unsigned ThisWords = NumBitWords(size()); 380 unsigned RHSWords = NumBitWords(RHS.size()); 381 unsigned i; 382 for (i = 0; i != std::min(ThisWords, RHSWords); ++i) 383 Bits[i] &= RHS.Bits[i]; 384 385 // Any bits that are just in this bitvector become zero, because they aren't 386 // in the RHS bit vector. Any words only in RHS are ignored because they 387 // are already zero in the LHS. 388 for (; i != ThisWords; ++i) 389 Bits[i] = 0; 390 391 return *this; 392 } 393 394 /// reset - Reset bits that are set in RHS. Same as *this &= ~RHS. 395 BitVector &reset(const BitVector &RHS) { 396 unsigned ThisWords = NumBitWords(size()); 397 unsigned RHSWords = NumBitWords(RHS.size()); 398 unsigned i; 399 for (i = 0; i != std::min(ThisWords, RHSWords); ++i) 400 Bits[i] &= ~RHS.Bits[i]; 401 return *this; 402 } 403 404 /// test - Check if (This - RHS) is zero. 405 /// This is the same as reset(RHS) and any(). 406 bool test(const BitVector &RHS) const { 407 unsigned ThisWords = NumBitWords(size()); 408 unsigned RHSWords = NumBitWords(RHS.size()); 409 unsigned i; 410 for (i = 0; i != std::min(ThisWords, RHSWords); ++i) 411 if ((Bits[i] & ~RHS.Bits[i]) != 0) 412 return true; 413 414 for (; i != ThisWords ; ++i) 415 if (Bits[i] != 0) 416 return true; 417 418 return false; 419 } 420 421 BitVector &operator|=(const BitVector &RHS) { 422 if (size() < RHS.size()) 423 resize(RHS.size()); 424 for (size_t i = 0, e = NumBitWords(RHS.size()); i != e; ++i) 425 Bits[i] |= RHS.Bits[i]; 426 return *this; 427 } 428 429 BitVector &operator^=(const BitVector &RHS) { 430 if (size() < RHS.size()) 431 resize(RHS.size()); 432 for (size_t i = 0, e = NumBitWords(RHS.size()); i != e; ++i) 433 Bits[i] ^= RHS.Bits[i]; 434 return *this; 435 } 436 437 // Assignment operator. 438 const BitVector &operator=(const BitVector &RHS) { 439 if (this == &RHS) return *this; 440 441 Size = RHS.size(); 442 unsigned RHSWords = NumBitWords(Size); 443 if (Size <= Capacity * BITWORD_SIZE) { 444 if (Size) 445 std::memcpy(Bits, RHS.Bits, RHSWords * sizeof(BitWord)); 446 clear_unused_bits(); 447 return *this; 448 } 449 450 // Grow the bitvector to have enough elements. 451 Capacity = RHSWords; 452 BitWord *NewBits = (BitWord *)std::malloc(Capacity * sizeof(BitWord)); 453 std::memcpy(NewBits, RHS.Bits, Capacity * sizeof(BitWord)); 454 455 // Destroy the old bits. 456 std::free(Bits); 457 Bits = NewBits; 458 459 return *this; 460 } 461 462 const BitVector &operator=(BitVector &&RHS) { 463 if (this == &RHS) return *this; 464 465 std::free(Bits); 466 Bits = RHS.Bits; 467 Size = RHS.Size; 468 Capacity = RHS.Capacity; 469 470 RHS.Bits = nullptr; 471 472 return *this; 473 } 474 475 void swap(BitVector &RHS) { 476 std::swap(Bits, RHS.Bits); 477 std::swap(Size, RHS.Size); 478 std::swap(Capacity, RHS.Capacity); 479 } 480 481 //===--------------------------------------------------------------------===// 482 // Portable bit mask operations. 483 //===--------------------------------------------------------------------===// 484 // 485 // These methods all operate on arrays of uint32_t, each holding 32 bits. The 486 // fixed word size makes it easier to work with literal bit vector constants 487 // in portable code. 488 // 489 // The LSB in each word is the lowest numbered bit. The size of a portable 490 // bit mask is always a whole multiple of 32 bits. If no bit mask size is 491 // given, the bit mask is assumed to cover the entire BitVector. 492 493 /// setBitsInMask - Add '1' bits from Mask to this vector. Don't resize. 494 /// This computes "*this |= Mask". 495 void setBitsInMask(const uint32_t *Mask, unsigned MaskWords = ~0u) { 496 applyMask<true, false>(Mask, MaskWords); 497 } 498 499 /// clearBitsInMask - Clear any bits in this vector that are set in Mask. 500 /// Don't resize. This computes "*this &= ~Mask". 501 void clearBitsInMask(const uint32_t *Mask, unsigned MaskWords = ~0u) { 502 applyMask<false, false>(Mask, MaskWords); 503 } 504 505 /// setBitsNotInMask - Add a bit to this vector for every '0' bit in Mask. 506 /// Don't resize. This computes "*this |= ~Mask". 507 void setBitsNotInMask(const uint32_t *Mask, unsigned MaskWords = ~0u) { 508 applyMask<true, true>(Mask, MaskWords); 509 } 510 511 /// clearBitsNotInMask - Clear a bit in this vector for every '0' bit in Mask. 512 /// Don't resize. This computes "*this &= Mask". 513 void clearBitsNotInMask(const uint32_t *Mask, unsigned MaskWords = ~0u) { 514 applyMask<false, true>(Mask, MaskWords); 515 } 516 517private: 518 unsigned NumBitWords(unsigned S) const { 519 return (S + BITWORD_SIZE-1) / BITWORD_SIZE; 520 } 521 522 // Set the unused bits in the high words. 523 void set_unused_bits(bool t = true) { 524 // Set high words first. 525 unsigned UsedWords = NumBitWords(Size); 526 if (Capacity > UsedWords) 527 init_words(&Bits[UsedWords], (Capacity-UsedWords), t); 528 529 // Then set any stray high bits of the last used word. 530 unsigned ExtraBits = Size % BITWORD_SIZE; 531 if (ExtraBits) { 532 BitWord ExtraBitMask = ~0UL << ExtraBits; 533 if (t) 534 Bits[UsedWords-1] |= ExtraBitMask; 535 else 536 Bits[UsedWords-1] &= ~ExtraBitMask; 537 } 538 } 539 540 // Clear the unused bits in the high words. 541 void clear_unused_bits() { 542 set_unused_bits(false); 543 } 544 545 void grow(unsigned NewSize) { 546 Capacity = std::max(NumBitWords(NewSize), Capacity * 2); 547 Bits = (BitWord *)std::realloc(Bits, Capacity * sizeof(BitWord)); 548 549 clear_unused_bits(); 550 } 551 552 void init_words(BitWord *B, unsigned NumWords, bool t) { 553 memset(B, 0 - (int)t, NumWords*sizeof(BitWord)); 554 } 555 556 template<bool AddBits, bool InvertMask> 557 void applyMask(const uint32_t *Mask, unsigned MaskWords) { 558 assert(BITWORD_SIZE % 32 == 0 && "Unsupported BitWord size."); 559 MaskWords = std::min(MaskWords, (size() + 31) / 32); 560 const unsigned Scale = BITWORD_SIZE / 32; 561 unsigned i; 562 for (i = 0; MaskWords >= Scale; ++i, MaskWords -= Scale) { 563 BitWord BW = Bits[i]; 564 // This inner loop should unroll completely when BITWORD_SIZE > 32. 565 for (unsigned b = 0; b != BITWORD_SIZE; b += 32) { 566 uint32_t M = *Mask++; 567 if (InvertMask) M = ~M; 568 if (AddBits) BW |= BitWord(M) << b; 569 else BW &= ~(BitWord(M) << b); 570 } 571 Bits[i] = BW; 572 } 573 for (unsigned b = 0; MaskWords; b += 32, --MaskWords) { 574 uint32_t M = *Mask++; 575 if (InvertMask) M = ~M; 576 if (AddBits) Bits[i] |= BitWord(M) << b; 577 else Bits[i] &= ~(BitWord(M) << b); 578 } 579 if (AddBits) 580 clear_unused_bits(); 581 } 582}; 583 584} // End llvm namespace 585 586namespace std { 587 /// Implement std::swap in terms of BitVector swap. 588 inline void 589 swap(llvm::BitVector &LHS, llvm::BitVector &RHS) { 590 LHS.swap(RHS); 591 } 592} 593 594#endif 595