1 //===-- llvm/ADT/Hashing.h - Utilities for hashing --------------*- 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 // This file implements the newly proposed standard C++ interfaces for hashing 10 // arbitrary data and building hash functions for user-defined types. This 11 // interface was originally proposed in N3333[1] and is currently under review 12 // for inclusion in a future TR and/or standard. 13 // 14 // The primary interfaces provide are comprised of one type and three functions: 15 // 16 // -- 'hash_code' class is an opaque type representing the hash code for some 17 // data. It is the intended product of hashing, and can be used to implement 18 // hash tables, checksumming, and other common uses of hashes. It is not an 19 // integer type (although it can be converted to one) because it is risky 20 // to assume much about the internals of a hash_code. In particular, each 21 // execution of the program has a high probability of producing a different 22 // hash_code for a given input. Thus their values are not stable to save or 23 // persist, and should only be used during the execution for the 24 // construction of hashing datastructures. 25 // 26 // -- 'hash_value' is a function designed to be overloaded for each 27 // user-defined type which wishes to be used within a hashing context. It 28 // should be overloaded within the user-defined type's namespace and found 29 // via ADL. Overloads for primitive types are provided by this library. 30 // 31 // -- 'hash_combine' and 'hash_combine_range' are functions designed to aid 32 // programmers in easily and intuitively combining a set of data into 33 // a single hash_code for their object. They should only logically be used 34 // within the implementation of a 'hash_value' routine or similar context. 35 // 36 // Note that 'hash_combine_range' contains very special logic for hashing 37 // a contiguous array of integers or pointers. This logic is *extremely* fast, 38 // on a modern Intel "Gainestown" Xeon (Nehalem uarch) @2.2 GHz, these were 39 // benchmarked at over 6.5 GiB/s for large keys, and <20 cycles/hash for keys 40 // under 32-bytes. 41 // 42 //===----------------------------------------------------------------------===// 43 44 #ifndef LLVM_ADT_HASHING_H 45 #define LLVM_ADT_HASHING_H 46 47 #include "llvm/Support/DataTypes.h" 48 #include "llvm/Support/ErrorHandling.h" 49 #include "llvm/Support/SwapByteOrder.h" 50 #include "llvm/Support/type_traits.h" 51 #include <algorithm> 52 #include <cassert> 53 #include <cstring> 54 #include <string> 55 #include <tuple> 56 #include <utility> 57 58 namespace llvm { 59 template <typename T, typename Enable> struct DenseMapInfo; 60 61 /// An opaque object representing a hash code. 62 /// 63 /// This object represents the result of hashing some entity. It is intended to 64 /// be used to implement hashtables or other hashing-based data structures. 65 /// While it wraps and exposes a numeric value, this value should not be 66 /// trusted to be stable or predictable across processes or executions. 67 /// 68 /// In order to obtain the hash_code for an object 'x': 69 /// \code 70 /// using llvm::hash_value; 71 /// llvm::hash_code code = hash_value(x); 72 /// \endcode 73 class hash_code { 74 size_t value; 75 76 public: 77 /// Default construct a hash_code. 78 /// Note that this leaves the value uninitialized. 79 hash_code() = default; 80 81 /// Form a hash code directly from a numerical value. 82 hash_code(size_t value) : value(value) {} 83 84 /// Convert the hash code to its numerical value for use. 85 /*explicit*/ operator size_t() const { return value; } 86 87 friend bool operator==(const hash_code &lhs, const hash_code &rhs) { 88 return lhs.value == rhs.value; 89 } 90 friend bool operator!=(const hash_code &lhs, const hash_code &rhs) { 91 return lhs.value != rhs.value; 92 } 93 94 /// Allow a hash_code to be directly run through hash_value. 95 friend size_t hash_value(const hash_code &code) { return code.value; } 96 }; 97 98 /// Compute a hash_code for any integer value. 99 /// 100 /// Note that this function is intended to compute the same hash_code for 101 /// a particular value without regard to the pre-promotion type. This is in 102 /// contrast to hash_combine which may produce different hash_codes for 103 /// differing argument types even if they would implicit promote to a common 104 /// type without changing the value. 105 template <typename T> 106 std::enable_if_t<is_integral_or_enum<T>::value, hash_code> hash_value(T value); 107 108 /// Compute a hash_code for a pointer's address. 109 /// 110 /// N.B.: This hashes the *address*. Not the value and not the type. 111 template <typename T> hash_code hash_value(const T *ptr); 112 113 /// Compute a hash_code for a pair of objects. 114 template <typename T, typename U> 115 hash_code hash_value(const std::pair<T, U> &arg); 116 117 /// Compute a hash_code for a tuple. 118 template <typename... Ts> 119 hash_code hash_value(const std::tuple<Ts...> &arg); 120 121 /// Compute a hash_code for a standard string. 122 template <typename T> 123 hash_code hash_value(const std::basic_string<T> &arg); 124 125 126 /// Override the execution seed with a fixed value. 127 /// 128 /// This hashing library uses a per-execution seed designed to change on each 129 /// run with high probability in order to ensure that the hash codes are not 130 /// attackable and to ensure that output which is intended to be stable does 131 /// not rely on the particulars of the hash codes produced. 132 /// 133 /// That said, there are use cases where it is important to be able to 134 /// reproduce *exactly* a specific behavior. To that end, we provide a function 135 /// which will forcibly set the seed to a fixed value. This must be done at the 136 /// start of the program, before any hashes are computed. Also, it cannot be 137 /// undone. This makes it thread-hostile and very hard to use outside of 138 /// immediately on start of a simple program designed for reproducible 139 /// behavior. 140 void set_fixed_execution_hash_seed(uint64_t fixed_value); 141 142 143 // All of the implementation details of actually computing the various hash 144 // code values are held within this namespace. These routines are included in 145 // the header file mainly to allow inlining and constant propagation. 146 namespace hashing { 147 namespace detail { 148 149 inline uint64_t fetch64(const char *p) { 150 uint64_t result; 151 memcpy(&result, p, sizeof(result)); 152 if (sys::IsBigEndianHost) 153 sys::swapByteOrder(result); 154 return result; 155 } 156 157 inline uint32_t fetch32(const char *p) { 158 uint32_t result; 159 memcpy(&result, p, sizeof(result)); 160 if (sys::IsBigEndianHost) 161 sys::swapByteOrder(result); 162 return result; 163 } 164 165 /// Some primes between 2^63 and 2^64 for various uses. 166 static constexpr uint64_t k0 = 0xc3a5c85c97cb3127ULL; 167 static constexpr uint64_t k1 = 0xb492b66fbe98f273ULL; 168 static constexpr uint64_t k2 = 0x9ae16a3b2f90404fULL; 169 static constexpr uint64_t k3 = 0xc949d7c7509e6557ULL; 170 171 /// Bitwise right rotate. 172 /// Normally this will compile to a single instruction, especially if the 173 /// shift is a manifest constant. 174 inline uint64_t rotate(uint64_t val, size_t shift) { 175 // Avoid shifting by 64: doing so yields an undefined result. 176 return shift == 0 ? val : ((val >> shift) | (val << (64 - shift))); 177 } 178 179 inline uint64_t shift_mix(uint64_t val) { 180 return val ^ (val >> 47); 181 } 182 183 inline uint64_t hash_16_bytes(uint64_t low, uint64_t high) { 184 // Murmur-inspired hashing. 185 const uint64_t kMul = 0x9ddfea08eb382d69ULL; 186 uint64_t a = (low ^ high) * kMul; 187 a ^= (a >> 47); 188 uint64_t b = (high ^ a) * kMul; 189 b ^= (b >> 47); 190 b *= kMul; 191 return b; 192 } 193 194 inline uint64_t hash_1to3_bytes(const char *s, size_t len, uint64_t seed) { 195 uint8_t a = s[0]; 196 uint8_t b = s[len >> 1]; 197 uint8_t c = s[len - 1]; 198 uint32_t y = static_cast<uint32_t>(a) + (static_cast<uint32_t>(b) << 8); 199 uint32_t z = static_cast<uint32_t>(len) + (static_cast<uint32_t>(c) << 2); 200 return shift_mix(y * k2 ^ z * k3 ^ seed) * k2; 201 } 202 203 inline uint64_t hash_4to8_bytes(const char *s, size_t len, uint64_t seed) { 204 uint64_t a = fetch32(s); 205 return hash_16_bytes(len + (a << 3), seed ^ fetch32(s + len - 4)); 206 } 207 208 inline uint64_t hash_9to16_bytes(const char *s, size_t len, uint64_t seed) { 209 uint64_t a = fetch64(s); 210 uint64_t b = fetch64(s + len - 8); 211 return hash_16_bytes(seed ^ a, rotate(b + len, len)) ^ b; 212 } 213 214 inline uint64_t hash_17to32_bytes(const char *s, size_t len, uint64_t seed) { 215 uint64_t a = fetch64(s) * k1; 216 uint64_t b = fetch64(s + 8); 217 uint64_t c = fetch64(s + len - 8) * k2; 218 uint64_t d = fetch64(s + len - 16) * k0; 219 return hash_16_bytes(rotate(a - b, 43) + rotate(c ^ seed, 30) + d, 220 a + rotate(b ^ k3, 20) - c + len + seed); 221 } 222 223 inline uint64_t hash_33to64_bytes(const char *s, size_t len, uint64_t seed) { 224 uint64_t z = fetch64(s + 24); 225 uint64_t a = fetch64(s) + (len + fetch64(s + len - 16)) * k0; 226 uint64_t b = rotate(a + z, 52); 227 uint64_t c = rotate(a, 37); 228 a += fetch64(s + 8); 229 c += rotate(a, 7); 230 a += fetch64(s + 16); 231 uint64_t vf = a + z; 232 uint64_t vs = b + rotate(a, 31) + c; 233 a = fetch64(s + 16) + fetch64(s + len - 32); 234 z = fetch64(s + len - 8); 235 b = rotate(a + z, 52); 236 c = rotate(a, 37); 237 a += fetch64(s + len - 24); 238 c += rotate(a, 7); 239 a += fetch64(s + len - 16); 240 uint64_t wf = a + z; 241 uint64_t ws = b + rotate(a, 31) + c; 242 uint64_t r = shift_mix((vf + ws) * k2 + (wf + vs) * k0); 243 return shift_mix((seed ^ (r * k0)) + vs) * k2; 244 } 245 246 inline uint64_t hash_short(const char *s, size_t length, uint64_t seed) { 247 if (length >= 4 && length <= 8) 248 return hash_4to8_bytes(s, length, seed); 249 if (length > 8 && length <= 16) 250 return hash_9to16_bytes(s, length, seed); 251 if (length > 16 && length <= 32) 252 return hash_17to32_bytes(s, length, seed); 253 if (length > 32) 254 return hash_33to64_bytes(s, length, seed); 255 if (length != 0) 256 return hash_1to3_bytes(s, length, seed); 257 258 return k2 ^ seed; 259 } 260 261 /// The intermediate state used during hashing. 262 /// Currently, the algorithm for computing hash codes is based on CityHash and 263 /// keeps 56 bytes of arbitrary state. 264 struct hash_state { 265 uint64_t h0 = 0, h1 = 0, h2 = 0, h3 = 0, h4 = 0, h5 = 0, h6 = 0; 266 267 /// Create a new hash_state structure and initialize it based on the 268 /// seed and the first 64-byte chunk. 269 /// This effectively performs the initial mix. 270 static hash_state create(const char *s, uint64_t seed) { 271 hash_state state = { 272 0, seed, hash_16_bytes(seed, k1), rotate(seed ^ k1, 49), 273 seed * k1, shift_mix(seed), 0 }; 274 state.h6 = hash_16_bytes(state.h4, state.h5); 275 state.mix(s); 276 return state; 277 } 278 279 /// Mix 32-bytes from the input sequence into the 16-bytes of 'a' 280 /// and 'b', including whatever is already in 'a' and 'b'. 281 static void mix_32_bytes(const char *s, uint64_t &a, uint64_t &b) { 282 a += fetch64(s); 283 uint64_t c = fetch64(s + 24); 284 b = rotate(b + a + c, 21); 285 uint64_t d = a; 286 a += fetch64(s + 8) + fetch64(s + 16); 287 b += rotate(a, 44) + d; 288 a += c; 289 } 290 291 /// Mix in a 64-byte buffer of data. 292 /// We mix all 64 bytes even when the chunk length is smaller, but we 293 /// record the actual length. 294 void mix(const char *s) { 295 h0 = rotate(h0 + h1 + h3 + fetch64(s + 8), 37) * k1; 296 h1 = rotate(h1 + h4 + fetch64(s + 48), 42) * k1; 297 h0 ^= h6; 298 h1 += h3 + fetch64(s + 40); 299 h2 = rotate(h2 + h5, 33) * k1; 300 h3 = h4 * k1; 301 h4 = h0 + h5; 302 mix_32_bytes(s, h3, h4); 303 h5 = h2 + h6; 304 h6 = h1 + fetch64(s + 16); 305 mix_32_bytes(s + 32, h5, h6); 306 std::swap(h2, h0); 307 } 308 309 /// Compute the final 64-bit hash code value based on the current 310 /// state and the length of bytes hashed. 311 uint64_t finalize(size_t length) { 312 return hash_16_bytes(hash_16_bytes(h3, h5) + shift_mix(h1) * k1 + h2, 313 hash_16_bytes(h4, h6) + shift_mix(length) * k1 + h0); 314 } 315 }; 316 317 318 /// A global, fixed seed-override variable. 319 /// 320 /// This variable can be set using the \see llvm::set_fixed_execution_seed 321 /// function. See that function for details. Do not, under any circumstances, 322 /// set or read this variable. 323 extern uint64_t fixed_seed_override; 324 325 inline uint64_t get_execution_seed() { 326 // FIXME: This needs to be a per-execution seed. This is just a placeholder 327 // implementation. Switching to a per-execution seed is likely to flush out 328 // instability bugs and so will happen as its own commit. 329 // 330 // However, if there is a fixed seed override set the first time this is 331 // called, return that instead of the per-execution seed. 332 const uint64_t seed_prime = 0xff51afd7ed558ccdULL; 333 static uint64_t seed = fixed_seed_override ? fixed_seed_override : seed_prime; 334 return seed; 335 } 336 337 338 /// Trait to indicate whether a type's bits can be hashed directly. 339 /// 340 /// A type trait which is true if we want to combine values for hashing by 341 /// reading the underlying data. It is false if values of this type must 342 /// first be passed to hash_value, and the resulting hash_codes combined. 343 // 344 // FIXME: We want to replace is_integral_or_enum and is_pointer here with 345 // a predicate which asserts that comparing the underlying storage of two 346 // values of the type for equality is equivalent to comparing the two values 347 // for equality. For all the platforms we care about, this holds for integers 348 // and pointers, but there are platforms where it doesn't and we would like to 349 // support user-defined types which happen to satisfy this property. 350 template <typename T> struct is_hashable_data 351 : std::integral_constant<bool, ((is_integral_or_enum<T>::value || 352 std::is_pointer<T>::value) && 353 64 % sizeof(T) == 0)> {}; 354 355 // Special case std::pair to detect when both types are viable and when there 356 // is no alignment-derived padding in the pair. This is a bit of a lie because 357 // std::pair isn't truly POD, but it's close enough in all reasonable 358 // implementations for our use case of hashing the underlying data. 359 template <typename T, typename U> struct is_hashable_data<std::pair<T, U> > 360 : std::integral_constant<bool, (is_hashable_data<T>::value && 361 is_hashable_data<U>::value && 362 (sizeof(T) + sizeof(U)) == 363 sizeof(std::pair<T, U>))> {}; 364 365 /// Helper to get the hashable data representation for a type. 366 /// This variant is enabled when the type itself can be used. 367 template <typename T> 368 std::enable_if_t<is_hashable_data<T>::value, T> 369 get_hashable_data(const T &value) { 370 return value; 371 } 372 /// Helper to get the hashable data representation for a type. 373 /// This variant is enabled when we must first call hash_value and use the 374 /// result as our data. 375 template <typename T> 376 std::enable_if_t<!is_hashable_data<T>::value, size_t> 377 get_hashable_data(const T &value) { 378 using ::llvm::hash_value; 379 return hash_value(value); 380 } 381 382 /// Helper to store data from a value into a buffer and advance the 383 /// pointer into that buffer. 384 /// 385 /// This routine first checks whether there is enough space in the provided 386 /// buffer, and if not immediately returns false. If there is space, it 387 /// copies the underlying bytes of value into the buffer, advances the 388 /// buffer_ptr past the copied bytes, and returns true. 389 template <typename T> 390 bool store_and_advance(char *&buffer_ptr, char *buffer_end, const T& value, 391 size_t offset = 0) { 392 size_t store_size = sizeof(value) - offset; 393 if (buffer_ptr + store_size > buffer_end) 394 return false; 395 const char *value_data = reinterpret_cast<const char *>(&value); 396 memcpy(buffer_ptr, value_data + offset, store_size); 397 buffer_ptr += store_size; 398 return true; 399 } 400 401 /// Implement the combining of integral values into a hash_code. 402 /// 403 /// This overload is selected when the value type of the iterator is 404 /// integral. Rather than computing a hash_code for each object and then 405 /// combining them, this (as an optimization) directly combines the integers. 406 template <typename InputIteratorT> 407 hash_code hash_combine_range_impl(InputIteratorT first, InputIteratorT last) { 408 const uint64_t seed = get_execution_seed(); 409 char buffer[64], *buffer_ptr = buffer; 410 char *const buffer_end = std::end(buffer); 411 while (first != last && store_and_advance(buffer_ptr, buffer_end, 412 get_hashable_data(*first))) 413 ++first; 414 if (first == last) 415 return hash_short(buffer, buffer_ptr - buffer, seed); 416 assert(buffer_ptr == buffer_end); 417 418 hash_state state = state.create(buffer, seed); 419 size_t length = 64; 420 while (first != last) { 421 // Fill up the buffer. We don't clear it, which re-mixes the last round 422 // when only a partial 64-byte chunk is left. 423 buffer_ptr = buffer; 424 while (first != last && store_and_advance(buffer_ptr, buffer_end, 425 get_hashable_data(*first))) 426 ++first; 427 428 // Rotate the buffer if we did a partial fill in order to simulate doing 429 // a mix of the last 64-bytes. That is how the algorithm works when we 430 // have a contiguous byte sequence, and we want to emulate that here. 431 std::rotate(buffer, buffer_ptr, buffer_end); 432 433 // Mix this chunk into the current state. 434 state.mix(buffer); 435 length += buffer_ptr - buffer; 436 }; 437 438 return state.finalize(length); 439 } 440 441 /// Implement the combining of integral values into a hash_code. 442 /// 443 /// This overload is selected when the value type of the iterator is integral 444 /// and when the input iterator is actually a pointer. Rather than computing 445 /// a hash_code for each object and then combining them, this (as an 446 /// optimization) directly combines the integers. Also, because the integers 447 /// are stored in contiguous memory, this routine avoids copying each value 448 /// and directly reads from the underlying memory. 449 template <typename ValueT> 450 std::enable_if_t<is_hashable_data<ValueT>::value, hash_code> 451 hash_combine_range_impl(ValueT *first, ValueT *last) { 452 const uint64_t seed = get_execution_seed(); 453 const char *s_begin = reinterpret_cast<const char *>(first); 454 const char *s_end = reinterpret_cast<const char *>(last); 455 const size_t length = std::distance(s_begin, s_end); 456 if (length <= 64) 457 return hash_short(s_begin, length, seed); 458 459 const char *s_aligned_end = s_begin + (length & ~63); 460 hash_state state = state.create(s_begin, seed); 461 s_begin += 64; 462 while (s_begin != s_aligned_end) { 463 state.mix(s_begin); 464 s_begin += 64; 465 } 466 if (length & 63) 467 state.mix(s_end - 64); 468 469 return state.finalize(length); 470 } 471 472 } // namespace detail 473 } // namespace hashing 474 475 476 /// Compute a hash_code for a sequence of values. 477 /// 478 /// This hashes a sequence of values. It produces the same hash_code as 479 /// 'hash_combine(a, b, c, ...)', but can run over arbitrary sized sequences 480 /// and is significantly faster given pointers and types which can be hashed as 481 /// a sequence of bytes. 482 template <typename InputIteratorT> 483 hash_code hash_combine_range(InputIteratorT first, InputIteratorT last) { 484 return ::llvm::hashing::detail::hash_combine_range_impl(first, last); 485 } 486 487 488 // Implementation details for hash_combine. 489 namespace hashing { 490 namespace detail { 491 492 /// Helper class to manage the recursive combining of hash_combine 493 /// arguments. 494 /// 495 /// This class exists to manage the state and various calls involved in the 496 /// recursive combining of arguments used in hash_combine. It is particularly 497 /// useful at minimizing the code in the recursive calls to ease the pain 498 /// caused by a lack of variadic functions. 499 struct hash_combine_recursive_helper { 500 char buffer[64] = {}; 501 hash_state state; 502 const uint64_t seed; 503 504 public: 505 /// Construct a recursive hash combining helper. 506 /// 507 /// This sets up the state for a recursive hash combine, including getting 508 /// the seed and buffer setup. 509 hash_combine_recursive_helper() 510 : seed(get_execution_seed()) {} 511 512 /// Combine one chunk of data into the current in-flight hash. 513 /// 514 /// This merges one chunk of data into the hash. First it tries to buffer 515 /// the data. If the buffer is full, it hashes the buffer into its 516 /// hash_state, empties it, and then merges the new chunk in. This also 517 /// handles cases where the data straddles the end of the buffer. 518 template <typename T> 519 char *combine_data(size_t &length, char *buffer_ptr, char *buffer_end, T data) { 520 if (!store_and_advance(buffer_ptr, buffer_end, data)) { 521 // Check for skew which prevents the buffer from being packed, and do 522 // a partial store into the buffer to fill it. This is only a concern 523 // with the variadic combine because that formation can have varying 524 // argument types. 525 size_t partial_store_size = buffer_end - buffer_ptr; 526 memcpy(buffer_ptr, &data, partial_store_size); 527 528 // If the store fails, our buffer is full and ready to hash. We have to 529 // either initialize the hash state (on the first full buffer) or mix 530 // this buffer into the existing hash state. Length tracks the *hashed* 531 // length, not the buffered length. 532 if (length == 0) { 533 state = state.create(buffer, seed); 534 length = 64; 535 } else { 536 // Mix this chunk into the current state and bump length up by 64. 537 state.mix(buffer); 538 length += 64; 539 } 540 // Reset the buffer_ptr to the head of the buffer for the next chunk of 541 // data. 542 buffer_ptr = buffer; 543 544 // Try again to store into the buffer -- this cannot fail as we only 545 // store types smaller than the buffer. 546 if (!store_and_advance(buffer_ptr, buffer_end, data, 547 partial_store_size)) 548 llvm_unreachable("buffer smaller than stored type"); 549 } 550 return buffer_ptr; 551 } 552 553 /// Recursive, variadic combining method. 554 /// 555 /// This function recurses through each argument, combining that argument 556 /// into a single hash. 557 template <typename T, typename ...Ts> 558 hash_code combine(size_t length, char *buffer_ptr, char *buffer_end, 559 const T &arg, const Ts &...args) { 560 buffer_ptr = combine_data(length, buffer_ptr, buffer_end, get_hashable_data(arg)); 561 562 // Recurse to the next argument. 563 return combine(length, buffer_ptr, buffer_end, args...); 564 } 565 566 /// Base case for recursive, variadic combining. 567 /// 568 /// The base case when combining arguments recursively is reached when all 569 /// arguments have been handled. It flushes the remaining buffer and 570 /// constructs a hash_code. 571 hash_code combine(size_t length, char *buffer_ptr, char *buffer_end) { 572 // Check whether the entire set of values fit in the buffer. If so, we'll 573 // use the optimized short hashing routine and skip state entirely. 574 if (length == 0) 575 return hash_short(buffer, buffer_ptr - buffer, seed); 576 577 // Mix the final buffer, rotating it if we did a partial fill in order to 578 // simulate doing a mix of the last 64-bytes. That is how the algorithm 579 // works when we have a contiguous byte sequence, and we want to emulate 580 // that here. 581 std::rotate(buffer, buffer_ptr, buffer_end); 582 583 // Mix this chunk into the current state. 584 state.mix(buffer); 585 length += buffer_ptr - buffer; 586 587 return state.finalize(length); 588 } 589 }; 590 591 } // namespace detail 592 } // namespace hashing 593 594 /// Combine values into a single hash_code. 595 /// 596 /// This routine accepts a varying number of arguments of any type. It will 597 /// attempt to combine them into a single hash_code. For user-defined types it 598 /// attempts to call a \see hash_value overload (via ADL) for the type. For 599 /// integer and pointer types it directly combines their data into the 600 /// resulting hash_code. 601 /// 602 /// The result is suitable for returning from a user's hash_value 603 /// *implementation* for their user-defined type. Consumers of a type should 604 /// *not* call this routine, they should instead call 'hash_value'. 605 template <typename ...Ts> hash_code hash_combine(const Ts &...args) { 606 // Recursively hash each argument using a helper class. 607 ::llvm::hashing::detail::hash_combine_recursive_helper helper; 608 return helper.combine(0, helper.buffer, helper.buffer + 64, args...); 609 } 610 611 // Implementation details for implementations of hash_value overloads provided 612 // here. 613 namespace hashing { 614 namespace detail { 615 616 /// Helper to hash the value of a single integer. 617 /// 618 /// Overloads for smaller integer types are not provided to ensure consistent 619 /// behavior in the presence of integral promotions. Essentially, 620 /// "hash_value('4')" and "hash_value('0' + 4)" should be the same. 621 inline hash_code hash_integer_value(uint64_t value) { 622 // Similar to hash_4to8_bytes but using a seed instead of length. 623 const uint64_t seed = get_execution_seed(); 624 const char *s = reinterpret_cast<const char *>(&value); 625 const uint64_t a = fetch32(s); 626 return hash_16_bytes(seed + (a << 3), fetch32(s + 4)); 627 } 628 629 } // namespace detail 630 } // namespace hashing 631 632 // Declared and documented above, but defined here so that any of the hashing 633 // infrastructure is available. 634 template <typename T> 635 std::enable_if_t<is_integral_or_enum<T>::value, hash_code> hash_value(T value) { 636 return ::llvm::hashing::detail::hash_integer_value( 637 static_cast<uint64_t>(value)); 638 } 639 640 // Declared and documented above, but defined here so that any of the hashing 641 // infrastructure is available. 642 template <typename T> hash_code hash_value(const T *ptr) { 643 return ::llvm::hashing::detail::hash_integer_value( 644 reinterpret_cast<uintptr_t>(ptr)); 645 } 646 647 // Declared and documented above, but defined here so that any of the hashing 648 // infrastructure is available. 649 template <typename T, typename U> 650 hash_code hash_value(const std::pair<T, U> &arg) { 651 return hash_combine(arg.first, arg.second); 652 } 653 654 // Implementation details for the hash_value overload for std::tuple<...>(...). 655 namespace hashing { 656 namespace detail { 657 658 template <typename... Ts, std::size_t... Indices> 659 hash_code hash_value_tuple_helper(const std::tuple<Ts...> &arg, 660 std::index_sequence<Indices...>) { 661 return hash_combine(std::get<Indices>(arg)...); 662 } 663 664 } // namespace detail 665 } // namespace hashing 666 667 template <typename... Ts> 668 hash_code hash_value(const std::tuple<Ts...> &arg) { 669 // TODO: Use std::apply when LLVM starts using C++17. 670 return ::llvm::hashing::detail::hash_value_tuple_helper( 671 arg, typename std::index_sequence_for<Ts...>()); 672 } 673 674 // Declared and documented above, but defined here so that any of the hashing 675 // infrastructure is available. 676 template <typename T> 677 hash_code hash_value(const std::basic_string<T> &arg) { 678 return hash_combine_range(arg.begin(), arg.end()); 679 } 680 681 template <> struct DenseMapInfo<hash_code, void> { 682 static inline hash_code getEmptyKey() { return hash_code(-1); } 683 static inline hash_code getTombstoneKey() { return hash_code(-2); } 684 static unsigned getHashValue(hash_code val) { return val; } 685 static bool isEqual(hash_code LHS, hash_code RHS) { return LHS == RHS; } 686 }; 687 688 } // namespace llvm 689 690 #endif 691