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