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