1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
|
// Copyright 2018 The Abseil Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//
// -----------------------------------------------------------------------------
// File: hash.h
// -----------------------------------------------------------------------------
//
#ifndef ABSL_HASH_INTERNAL_HASH_H_
#define ABSL_HASH_INTERNAL_HASH_H_
#include <algorithm>
#include <array>
#include <cmath>
#include <cstring>
#include <deque>
#include <forward_list>
#include <functional>
#include <iterator>
#include <limits>
#include <list>
#include <map>
#include <memory>
#include <set>
#include <string>
#include <tuple>
#include <type_traits>
#include <utility>
#include <vector>
#include "absl/base/internal/endian.h"
#include "absl/base/port.h"
#include "absl/container/fixed_array.h"
#include "absl/meta/type_traits.h"
#include "absl/numeric/int128.h"
#include "absl/strings/string_view.h"
#include "absl/types/optional.h"
#include "absl/types/variant.h"
#include "absl/utility/utility.h"
#include "absl/hash/internal/city.h"
namespace absl {
ABSL_NAMESPACE_BEGIN
namespace hash_internal {
// Internal detail: Large buffers are hashed in smaller chunks. This function
// returns the size of these chunks.
constexpr size_t PiecewiseChunkSize() { return 1024; }
// PiecewiseCombiner
//
// PiecewiseCombiner is an internal-only helper class for hashing a piecewise
// buffer of `char` or `unsigned char` as though it were contiguous. This class
// provides two methods:
//
// H add_buffer(state, data, size)
// H finalize(state)
//
// `add_buffer` can be called zero or more times, followed by a single call to
// `finalize`. This will produce the same hash expansion as concatenating each
// buffer piece into a single contiguous buffer, and passing this to
// `H::combine_contiguous`.
//
// Example usage:
// PiecewiseCombiner combiner;
// for (const auto& piece : pieces) {
// state = combiner.add_buffer(std::move(state), piece.data, piece.size);
// }
// return combiner.finalize(std::move(state));
class PiecewiseCombiner {
public:
PiecewiseCombiner() : position_(0) {}
PiecewiseCombiner(const PiecewiseCombiner&) = delete;
PiecewiseCombiner& operator=(const PiecewiseCombiner&) = delete;
// PiecewiseCombiner::add_buffer()
//
// Appends the given range of bytes to the sequence to be hashed, which may
// modify the provided hash state.
template <typename H>
H add_buffer(H state, const unsigned char* data, size_t size);
template <typename H>
H add_buffer(H state, const char* data, size_t size) {
return add_buffer(std::move(state),
reinterpret_cast<const unsigned char*>(data), size);
}
// PiecewiseCombiner::finalize()
//
// Finishes combining the hash sequence, which may may modify the provided
// hash state.
//
// Once finalize() is called, add_buffer() may no longer be called. The
// resulting hash state will be the same as if the pieces passed to
// add_buffer() were concatenated into a single flat buffer, and then provided
// to H::combine_contiguous().
template <typename H>
H finalize(H state);
private:
unsigned char buf_[PiecewiseChunkSize()];
size_t position_;
};
// HashStateBase
//
// A hash state object represents an intermediate state in the computation
// of an unspecified hash algorithm. `HashStateBase` provides a CRTP style
// base class for hash state implementations. Developers adding type support
// for `absl::Hash` should not rely on any parts of the state object other than
// the following member functions:
//
// * HashStateBase::combine()
// * HashStateBase::combine_contiguous()
//
// A derived hash state class of type `H` must provide a static member function
// with a signature similar to the following:
//
// `static H combine_contiguous(H state, const unsigned char*, size_t)`.
//
// `HashStateBase` will provide a complete implementation for a hash state
// object in terms of this method.
//
// Example:
//
// // Use CRTP to define your derived class.
// struct MyHashState : HashStateBase<MyHashState> {
// static H combine_contiguous(H state, const unsigned char*, size_t);
// using MyHashState::HashStateBase::combine;
// using MyHashState::HashStateBase::combine_contiguous;
// };
template <typename H>
class HashStateBase {
public:
// HashStateBase::combine()
//
// Combines an arbitrary number of values into a hash state, returning the
// updated state.
//
// Each of the value types `T` must be separately hashable by the Abseil
// hashing framework.
//
// NOTE:
//
// state = H::combine(std::move(state), value1, value2, value3);
//
// is guaranteed to produce the same hash expansion as:
//
// state = H::combine(std::move(state), value1);
// state = H::combine(std::move(state), value2);
// state = H::combine(std::move(state), value3);
template <typename T, typename... Ts>
static H combine(H state, const T& value, const Ts&... values);
static H combine(H state) { return state; }
// HashStateBase::combine_contiguous()
//
// Combines a contiguous array of `size` elements into a hash state, returning
// the updated state.
//
// NOTE:
//
// state = H::combine_contiguous(std::move(state), data, size);
//
// is NOT guaranteed to produce the same hash expansion as a for-loop (it may
// perform internal optimizations). If you need this guarantee, use the
// for-loop instead.
template <typename T>
static H combine_contiguous(H state, const T* data, size_t size);
using AbslInternalPiecewiseCombiner = PiecewiseCombiner;
};
// is_uniquely_represented
//
// `is_uniquely_represented<T>` is a trait class that indicates whether `T`
// is uniquely represented.
//
// A type is "uniquely represented" if two equal values of that type are
// guaranteed to have the same bytes in their underlying storage. In other
// words, if `a == b`, then `memcmp(&a, &b, sizeof(T))` is guaranteed to be
// zero. This property cannot be detected automatically, so this trait is false
// by default, but can be specialized by types that wish to assert that they are
// uniquely represented. This makes them eligible for certain optimizations.
//
// If you have any doubt whatsoever, do not specialize this template.
// The default is completely safe, and merely disables some optimizations
// that will not matter for most types. Specializing this template,
// on the other hand, can be very hazardous.
//
// To be uniquely represented, a type must not have multiple ways of
// representing the same value; for example, float and double are not
// uniquely represented, because they have distinct representations for
// +0 and -0. Furthermore, the type's byte representation must consist
// solely of user-controlled data, with no padding bits and no compiler-
// controlled data such as vptrs or sanitizer metadata. This is usually
// very difficult to guarantee, because in most cases the compiler can
// insert data and padding bits at its own discretion.
//
// If you specialize this template for a type `T`, you must do so in the file
// that defines that type (or in this file). If you define that specialization
// anywhere else, `is_uniquely_represented<T>` could have different meanings
// in different places.
//
// The Enable parameter is meaningless; it is provided as a convenience,
// to support certain SFINAE techniques when defining specializations.
template <typename T, typename Enable = void>
struct is_uniquely_represented : std::false_type {};
// is_uniquely_represented<unsigned char>
//
// unsigned char is a synonym for "byte", so it is guaranteed to be
// uniquely represented.
template <>
struct is_uniquely_represented<unsigned char> : std::true_type {};
// is_uniquely_represented for non-standard integral types
//
// Integral types other than bool should be uniquely represented on any
// platform that this will plausibly be ported to.
template <typename Integral>
struct is_uniquely_represented<
Integral, typename std::enable_if<std::is_integral<Integral>::value>::type>
: std::true_type {};
// is_uniquely_represented<bool>
//
//
template <>
struct is_uniquely_represented<bool> : std::false_type {};
// hash_bytes()
//
// Convenience function that combines `hash_state` with the byte representation
// of `value`.
template <typename H, typename T>
H hash_bytes(H hash_state, const T& value) {
const unsigned char* start = reinterpret_cast<const unsigned char*>(&value);
return H::combine_contiguous(std::move(hash_state), start, sizeof(value));
}
// -----------------------------------------------------------------------------
// AbslHashValue for Basic Types
// -----------------------------------------------------------------------------
// Note: Default `AbslHashValue` implementations live in `hash_internal`. This
// allows us to block lexical scope lookup when doing an unqualified call to
// `AbslHashValue` below. User-defined implementations of `AbslHashValue` can
// only be found via ADL.
// AbslHashValue() for hashing bool values
//
// We use SFINAE to ensure that this overload only accepts bool, not types that
// are convertible to bool.
template <typename H, typename B>
typename std::enable_if<std::is_same<B, bool>::value, H>::type AbslHashValue(
H hash_state, B value) {
return H::combine(std::move(hash_state),
static_cast<unsigned char>(value ? 1 : 0));
}
// AbslHashValue() for hashing enum values
template <typename H, typename Enum>
typename std::enable_if<std::is_enum<Enum>::value, H>::type AbslHashValue(
H hash_state, Enum e) {
// In practice, we could almost certainly just invoke hash_bytes directly,
// but it's possible that a sanitizer might one day want to
// store data in the unused bits of an enum. To avoid that risk, we
// convert to the underlying type before hashing. Hopefully this will get
// optimized away; if not, we can reopen discussion with c-toolchain-team.
return H::combine(std::move(hash_state),
static_cast<typename std::underlying_type<Enum>::type>(e));
}
// AbslHashValue() for hashing floating-point values
template <typename H, typename Float>
typename std::enable_if<std::is_same<Float, float>::value ||
std::is_same<Float, double>::value,
H>::type
AbslHashValue(H hash_state, Float value) {
return hash_internal::hash_bytes(std::move(hash_state),
value == 0 ? 0 : value);
}
// Long double has the property that it might have extra unused bytes in it.
// For example, in x86 sizeof(long double)==16 but it only really uses 80-bits
// of it. This means we can't use hash_bytes on a long double and have to
// convert it to something else first.
template <typename H, typename LongDouble>
typename std::enable_if<std::is_same<LongDouble, long double>::value, H>::type
AbslHashValue(H hash_state, LongDouble value) {
const int category = std::fpclassify(value);
switch (category) {
case FP_INFINITE:
// Add the sign bit to differentiate between +Inf and -Inf
hash_state = H::combine(std::move(hash_state), std::signbit(value));
break;
case FP_NAN:
case FP_ZERO:
default:
// Category is enough for these.
break;
case FP_NORMAL:
case FP_SUBNORMAL:
// We can't convert `value` directly to double because this would have
// undefined behavior if the value is out of range.
// std::frexp gives us a value in the range (-1, -.5] or [.5, 1) that is
// guaranteed to be in range for `double`. The truncation is
// implementation defined, but that works as long as it is deterministic.
int exp;
auto mantissa = static_cast<double>(std::frexp(value, &exp));
hash_state = H::combine(std::move(hash_state), mantissa, exp);
}
return H::combine(std::move(hash_state), category);
}
// AbslHashValue() for hashing pointers
template <typename H, typename T>
H AbslHashValue(H hash_state, T* ptr) {
auto v = reinterpret_cast<uintptr_t>(ptr);
// Due to alignment, pointers tend to have low bits as zero, and the next few
// bits follow a pattern since they are also multiples of some base value.
// Mixing the pointer twice helps prevent stuck low bits for certain alignment
// values.
return H::combine(std::move(hash_state), v, v);
}
// AbslHashValue() for hashing nullptr_t
template <typename H>
H AbslHashValue(H hash_state, std::nullptr_t) {
return H::combine(std::move(hash_state), static_cast<void*>(nullptr));
}
// -----------------------------------------------------------------------------
// AbslHashValue for Composite Types
// -----------------------------------------------------------------------------
// is_hashable()
//
// Trait class which returns true if T is hashable by the absl::Hash framework.
// Used for the AbslHashValue implementations for composite types below.
template <typename T>
struct is_hashable;
// AbslHashValue() for hashing pairs
template <typename H, typename T1, typename T2>
typename std::enable_if<is_hashable<T1>::value && is_hashable<T2>::value,
H>::type
AbslHashValue(H hash_state, const std::pair<T1, T2>& p) {
return H::combine(std::move(hash_state), p.first, p.second);
}
// hash_tuple()
//
// Helper function for hashing a tuple. The third argument should
// be an index_sequence running from 0 to tuple_size<Tuple> - 1.
template <typename H, typename Tuple, size_t... Is>
H hash_tuple(H hash_state, const Tuple& t, absl::index_sequence<Is...>) {
return H::combine(std::move(hash_state), std::get<Is>(t)...);
}
// AbslHashValue for hashing tuples
template <typename H, typename... Ts>
#if defined(_MSC_VER)
// This SFINAE gets MSVC confused under some conditions. Let's just disable it
// for now.
H
#else // _MSC_VER
typename std::enable_if<absl::conjunction<is_hashable<Ts>...>::value, H>::type
#endif // _MSC_VER
AbslHashValue(H hash_state, const std::tuple<Ts...>& t) {
return hash_internal::hash_tuple(std::move(hash_state), t,
absl::make_index_sequence<sizeof...(Ts)>());
}
// -----------------------------------------------------------------------------
// AbslHashValue for Pointers
// -----------------------------------------------------------------------------
// AbslHashValue for hashing unique_ptr
template <typename H, typename T, typename D>
H AbslHashValue(H hash_state, const std::unique_ptr<T, D>& ptr) {
return H::combine(std::move(hash_state), ptr.get());
}
// AbslHashValue for hashing shared_ptr
template <typename H, typename T>
H AbslHashValue(H hash_state, const std::shared_ptr<T>& ptr) {
return H::combine(std::move(hash_state), ptr.get());
}
// -----------------------------------------------------------------------------
// AbslHashValue for String-Like Types
// -----------------------------------------------------------------------------
// AbslHashValue for hashing strings
//
// All the string-like types supported here provide the same hash expansion for
// the same character sequence. These types are:
//
// - `absl::Cord`
// - `std::string` (and std::basic_string<char, std::char_traits<char>, A> for
// any allocator A)
// - `absl::string_view` and `std::string_view`
//
// For simplicity, we currently support only `char` strings. This support may
// be broadened, if necessary, but with some caution - this overload would
// misbehave in cases where the traits' `eq()` member isn't equivalent to `==`
// on the underlying character type.
template <typename H>
H AbslHashValue(H hash_state, absl::string_view str) {
return H::combine(
H::combine_contiguous(std::move(hash_state), str.data(), str.size()),
str.size());
}
// Support std::wstring, std::u16string and std::u32string.
template <typename Char, typename Alloc, typename H,
typename = absl::enable_if_t<std::is_same<Char, wchar_t>::value ||
std::is_same<Char, char16_t>::value ||
std::is_same<Char, char32_t>::value>>
H AbslHashValue(
H hash_state,
const std::basic_string<Char, std::char_traits<Char>, Alloc>& str) {
return H::combine(
H::combine_contiguous(std::move(hash_state), str.data(), str.size()),
str.size());
}
// -----------------------------------------------------------------------------
// AbslHashValue for Sequence Containers
// -----------------------------------------------------------------------------
// AbslHashValue for hashing std::array
template <typename H, typename T, size_t N>
typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
H hash_state, const std::array<T, N>& array) {
return H::combine_contiguous(std::move(hash_state), array.data(),
array.size());
}
// AbslHashValue for hashing std::deque
template <typename H, typename T, typename Allocator>
typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
H hash_state, const std::deque<T, Allocator>& deque) {
// TODO(gromer): investigate a more efficient implementation taking
// advantage of the chunk structure.
for (const auto& t : deque) {
hash_state = H::combine(std::move(hash_state), t);
}
return H::combine(std::move(hash_state), deque.size());
}
// AbslHashValue for hashing std::forward_list
template <typename H, typename T, typename Allocator>
typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
H hash_state, const std::forward_list<T, Allocator>& list) {
size_t size = 0;
for (const T& t : list) {
hash_state = H::combine(std::move(hash_state), t);
++size;
}
return H::combine(std::move(hash_state), size);
}
// AbslHashValue for hashing std::list
template <typename H, typename T, typename Allocator>
typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
H hash_state, const std::list<T, Allocator>& list) {
for (const auto& t : list) {
hash_state = H::combine(std::move(hash_state), t);
}
return H::combine(std::move(hash_state), list.size());
}
// AbslHashValue for hashing std::vector
//
// Do not use this for vector<bool>. It does not have a .data(), and a fallback
// for std::hash<> is most likely faster.
template <typename H, typename T, typename Allocator>
typename std::enable_if<is_hashable<T>::value && !std::is_same<T, bool>::value,
H>::type
AbslHashValue(H hash_state, const std::vector<T, Allocator>& vector) {
return H::combine(H::combine_contiguous(std::move(hash_state), vector.data(),
vector.size()),
vector.size());
}
// -----------------------------------------------------------------------------
// AbslHashValue for Ordered Associative Containers
// -----------------------------------------------------------------------------
// AbslHashValue for hashing std::map
template <typename H, typename Key, typename T, typename Compare,
typename Allocator>
typename std::enable_if<is_hashable<Key>::value && is_hashable<T>::value,
H>::type
AbslHashValue(H hash_state, const std::map<Key, T, Compare, Allocator>& map) {
for (const auto& t : map) {
hash_state = H::combine(std::move(hash_state), t);
}
return H::combine(std::move(hash_state), map.size());
}
// AbslHashValue for hashing std::multimap
template <typename H, typename Key, typename T, typename Compare,
typename Allocator>
typename std::enable_if<is_hashable<Key>::value && is_hashable<T>::value,
H>::type
AbslHashValue(H hash_state,
const std::multimap<Key, T, Compare, Allocator>& map) {
for (const auto& t : map) {
hash_state = H::combine(std::move(hash_state), t);
}
return H::combine(std::move(hash_state), map.size());
}
// AbslHashValue for hashing std::set
template <typename H, typename Key, typename Compare, typename Allocator>
typename std::enable_if<is_hashable<Key>::value, H>::type AbslHashValue(
H hash_state, const std::set<Key, Compare, Allocator>& set) {
for (const auto& t : set) {
hash_state = H::combine(std::move(hash_state), t);
}
return H::combine(std::move(hash_state), set.size());
}
// AbslHashValue for hashing std::multiset
template <typename H, typename Key, typename Compare, typename Allocator>
typename std::enable_if<is_hashable<Key>::value, H>::type AbslHashValue(
H hash_state, const std::multiset<Key, Compare, Allocator>& set) {
for (const auto& t : set) {
hash_state = H::combine(std::move(hash_state), t);
}
return H::combine(std::move(hash_state), set.size());
}
// -----------------------------------------------------------------------------
// AbslHashValue for Wrapper Types
// -----------------------------------------------------------------------------
// AbslHashValue for hashing std::reference_wrapper
template <typename H, typename T>
typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
H hash_state, std::reference_wrapper<T> opt) {
return H::combine(std::move(hash_state), opt.get());
}
// AbslHashValue for hashing absl::optional
template <typename H, typename T>
typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
H hash_state, const absl::optional<T>& opt) {
if (opt) hash_state = H::combine(std::move(hash_state), *opt);
return H::combine(std::move(hash_state), opt.has_value());
}
// VariantVisitor
template <typename H>
struct VariantVisitor {
H&& hash_state;
template <typename T>
H operator()(const T& t) const {
return H::combine(std::move(hash_state), t);
}
};
// AbslHashValue for hashing absl::variant
template <typename H, typename... T>
typename std::enable_if<conjunction<is_hashable<T>...>::value, H>::type
AbslHashValue(H hash_state, const absl::variant<T...>& v) {
if (!v.valueless_by_exception()) {
hash_state = absl::visit(VariantVisitor<H>{std::move(hash_state)}, v);
}
return H::combine(std::move(hash_state), v.index());
}
// -----------------------------------------------------------------------------
// AbslHashValue for Other Types
// -----------------------------------------------------------------------------
// AbslHashValue for hashing std::bitset is not defined, for the same reason as
// for vector<bool> (see std::vector above): It does not expose the raw bytes,
// and a fallback to std::hash<> is most likely faster.
// -----------------------------------------------------------------------------
// hash_range_or_bytes()
//
// Mixes all values in the range [data, data+size) into the hash state.
// This overload accepts only uniquely-represented types, and hashes them by
// hashing the entire range of bytes.
template <typename H, typename T>
typename std::enable_if<is_uniquely_represented<T>::value, H>::type
hash_range_or_bytes(H hash_state, const T* data, size_t size) {
const auto* bytes = reinterpret_cast<const unsigned char*>(data);
return H::combine_contiguous(std::move(hash_state), bytes, sizeof(T) * size);
}
// hash_range_or_bytes()
template <typename H, typename T>
typename std::enable_if<!is_uniquely_represented<T>::value, H>::type
hash_range_or_bytes(H hash_state, const T* data, size_t size) {
for (const auto end = data + size; data < end; ++data) {
hash_state = H::combine(std::move(hash_state), *data);
}
return hash_state;
}
#if defined(ABSL_INTERNAL_LEGACY_HASH_NAMESPACE) && \
ABSL_META_INTERNAL_STD_HASH_SFINAE_FRIENDLY_
#define ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ 1
#else
#define ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ 0
#endif
// HashSelect
//
// Type trait to select the appropriate hash implementation to use.
// HashSelect::type<T> will give the proper hash implementation, to be invoked
// as:
// HashSelect::type<T>::Invoke(state, value)
// Also, HashSelect::type<T>::value is a boolean equal to `true` if there is a
// valid `Invoke` function. Types that are not hashable will have a ::value of
// `false`.
struct HashSelect {
private:
struct State : HashStateBase<State> {
static State combine_contiguous(State hash_state, const unsigned char*,
size_t);
using State::HashStateBase::combine_contiguous;
};
struct UniquelyRepresentedProbe {
template <typename H, typename T>
static auto Invoke(H state, const T& value)
-> absl::enable_if_t<is_uniquely_represented<T>::value, H> {
return hash_internal::hash_bytes(std::move(state), value);
}
};
struct HashValueProbe {
template <typename H, typename T>
static auto Invoke(H state, const T& value) -> absl::enable_if_t<
std::is_same<H,
decltype(AbslHashValue(std::move(state), value))>::value,
H> {
return AbslHashValue(std::move(state), value);
}
};
struct LegacyHashProbe {
#if ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_
template <typename H, typename T>
static auto Invoke(H state, const T& value) -> absl::enable_if_t<
std::is_convertible<
decltype(ABSL_INTERNAL_LEGACY_HASH_NAMESPACE::hash<T>()(value)),
size_t>::value,
H> {
return hash_internal::hash_bytes(
std::move(state),
ABSL_INTERNAL_LEGACY_HASH_NAMESPACE::hash<T>{}(value));
}
#endif // ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_
};
struct StdHashProbe {
template <typename H, typename T>
static auto Invoke(H state, const T& value)
-> absl::enable_if_t<type_traits_internal::IsHashable<T>::value, H> {
return hash_internal::hash_bytes(std::move(state), std::hash<T>{}(value));
}
};
template <typename Hash, typename T>
struct Probe : Hash {
private:
template <typename H, typename = decltype(H::Invoke(
std::declval<State>(), std::declval<const T&>()))>
static std::true_type Test(int);
template <typename U>
static std::false_type Test(char);
public:
static constexpr bool value = decltype(Test<Hash>(0))::value;
};
public:
// Probe each implementation in order.
// disjunction provides short circuiting wrt instantiation.
template <typename T>
using Apply = absl::disjunction< //
Probe<UniquelyRepresentedProbe, T>, //
Probe<HashValueProbe, T>, //
Probe<LegacyHashProbe, T>, //
Probe<StdHashProbe, T>, //
std::false_type>;
};
template <typename T>
struct is_hashable
: std::integral_constant<bool, HashSelect::template Apply<T>::value> {};
// CityHashState
class ABSL_DLL CityHashState
: public HashStateBase<CityHashState> {
// absl::uint128 is not an alias or a thin wrapper around the intrinsic.
// We use the intrinsic when available to improve performance.
#ifdef ABSL_HAVE_INTRINSIC_INT128
using uint128 = __uint128_t;
#else // ABSL_HAVE_INTRINSIC_INT128
using uint128 = absl::uint128;
#endif // ABSL_HAVE_INTRINSIC_INT128
static constexpr uint64_t kMul =
sizeof(size_t) == 4 ? uint64_t{0xcc9e2d51}
: uint64_t{0x9ddfea08eb382d69};
template <typename T>
using IntegralFastPath =
conjunction<std::is_integral<T>, is_uniquely_represented<T>>;
public:
// Move only
CityHashState(CityHashState&&) = default;
CityHashState& operator=(CityHashState&&) = default;
// CityHashState::combine_contiguous()
//
// Fundamental base case for hash recursion: mixes the given range of bytes
// into the hash state.
static CityHashState combine_contiguous(CityHashState hash_state,
const unsigned char* first,
size_t size) {
return CityHashState(
CombineContiguousImpl(hash_state.state_, first, size,
std::integral_constant<int, sizeof(size_t)>{}));
}
using CityHashState::HashStateBase::combine_contiguous;
// CityHashState::hash()
//
// For performance reasons in non-opt mode, we specialize this for
// integral types.
// Otherwise we would be instantiating and calling dozens of functions for
// something that is just one multiplication and a couple xor's.
// The result should be the same as running the whole algorithm, but faster.
template <typename T, absl::enable_if_t<IntegralFastPath<T>::value, int> = 0>
static size_t hash(T value) {
return static_cast<size_t>(Mix(Seed(), static_cast<uint64_t>(value)));
}
// Overload of CityHashState::hash()
template <typename T, absl::enable_if_t<!IntegralFastPath<T>::value, int> = 0>
static size_t hash(const T& value) {
return static_cast<size_t>(combine(CityHashState{}, value).state_);
}
private:
// Invoked only once for a given argument; that plus the fact that this is
// move-only ensures that there is only one non-moved-from object.
CityHashState() : state_(Seed()) {}
// Workaround for MSVC bug.
// We make the type copyable to fix the calling convention, even though we
// never actually copy it. Keep it private to not affect the public API of the
// type.
CityHashState(const CityHashState&) = default;
explicit CityHashState(uint64_t state) : state_(state) {}
// Implementation of the base case for combine_contiguous where we actually
// mix the bytes into the state.
// Dispatch to different implementations of the combine_contiguous depending
// on the value of `sizeof(size_t)`.
static uint64_t CombineContiguousImpl(uint64_t state,
const unsigned char* first, size_t len,
std::integral_constant<int, 4>
/* sizeof_size_t */);
static uint64_t CombineContiguousImpl(uint64_t state,
const unsigned char* first, size_t len,
std::integral_constant<int, 8>
/* sizeof_size_t*/);
// Slow dispatch path for calls to CombineContiguousImpl with a size argument
// larger than PiecewiseChunkSize(). Has the same effect as calling
// CombineContiguousImpl() repeatedly with the chunk stride size.
static uint64_t CombineLargeContiguousImpl32(uint64_t state,
const unsigned char* first,
size_t len);
static uint64_t CombineLargeContiguousImpl64(uint64_t state,
const unsigned char* first,
size_t len);
// Reads 9 to 16 bytes from p.
// The first 8 bytes are in .first, the rest (zero padded) bytes are in
// .second.
static std::pair<uint64_t, uint64_t> Read9To16(const unsigned char* p,
size_t len) {
uint64_t high = little_endian::Load64(p + len - 8);
return {little_endian::Load64(p), high >> (128 - len * 8)};
}
// Reads 4 to 8 bytes from p. Zero pads to fill uint64_t.
static uint64_t Read4To8(const unsigned char* p, size_t len) {
return (static_cast<uint64_t>(little_endian::Load32(p + len - 4))
<< (len - 4) * 8) |
little_endian::Load32(p);
}
// Reads 1 to 3 bytes from p. Zero pads to fill uint32_t.
static uint32_t Read1To3(const unsigned char* p, size_t len) {
return static_cast<uint32_t>((p[0]) | //
(p[len / 2] << (len / 2 * 8)) | //
(p[len - 1] << ((len - 1) * 8)));
}
ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t Mix(uint64_t state, uint64_t v) {
using MultType =
absl::conditional_t<sizeof(size_t) == 4, uint64_t, uint128>;
// We do the addition in 64-bit space to make sure the 128-bit
// multiplication is fast. If we were to do it as MultType the compiler has
// to assume that the high word is non-zero and needs to perform 2
// multiplications instead of one.
MultType m = state + v;
m *= kMul;
return static_cast<uint64_t>(m ^ (m >> (sizeof(m) * 8 / 2)));
}
// Seed()
//
// A non-deterministic seed.
//
// The current purpose of this seed is to generate non-deterministic results
// and prevent having users depend on the particular hash values.
// It is not meant as a security feature right now, but it leaves the door
// open to upgrade it to a true per-process random seed. A true random seed
// costs more and we don't need to pay for that right now.
//
// On platforms with ASLR, we take advantage of it to make a per-process
// random value.
// See https://en.wikipedia.org/wiki/Address_space_layout_randomization
//
// On other platforms this is still going to be non-deterministic but most
// probably per-build and not per-process.
ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t Seed() {
return static_cast<uint64_t>(reinterpret_cast<uintptr_t>(kSeed));
}
static const void* const kSeed;
uint64_t state_;
};
// CityHashState::CombineContiguousImpl()
inline uint64_t CityHashState::CombineContiguousImpl(
uint64_t state, const unsigned char* first, size_t len,
std::integral_constant<int, 4> /* sizeof_size_t */) {
// For large values we use CityHash, for small ones we just use a
// multiplicative hash.
uint64_t v;
if (len > 8) {
if (ABSL_PREDICT_FALSE(len > PiecewiseChunkSize())) {
return CombineLargeContiguousImpl32(state, first, len);
}
v = absl::hash_internal::CityHash32(reinterpret_cast<const char*>(first), len);
} else if (len >= 4) {
v = Read4To8(first, len);
} else if (len > 0) {
v = Read1To3(first, len);
} else {
// Empty ranges have no effect.
return state;
}
return Mix(state, v);
}
// Overload of CityHashState::CombineContiguousImpl()
inline uint64_t CityHashState::CombineContiguousImpl(
uint64_t state, const unsigned char* first, size_t len,
std::integral_constant<int, 8> /* sizeof_size_t */) {
// For large values we use CityHash, for small ones we just use a
// multiplicative hash.
uint64_t v;
if (len > 16) {
if (ABSL_PREDICT_FALSE(len > PiecewiseChunkSize())) {
return CombineLargeContiguousImpl64(state, first, len);
}
v = absl::hash_internal::CityHash64(reinterpret_cast<const char*>(first), len);
} else if (len > 8) {
auto p = Read9To16(first, len);
state = Mix(state, p.first);
v = p.second;
} else if (len >= 4) {
v = Read4To8(first, len);
} else if (len > 0) {
v = Read1To3(first, len);
} else {
// Empty ranges have no effect.
return state;
}
return Mix(state, v);
}
struct AggregateBarrier {};
// HashImpl
// Add a private base class to make sure this type is not an aggregate.
// Aggregates can be aggregate initialized even if the default constructor is
// deleted.
struct PoisonedHash : private AggregateBarrier {
PoisonedHash() = delete;
PoisonedHash(const PoisonedHash&) = delete;
PoisonedHash& operator=(const PoisonedHash&) = delete;
};
template <typename T>
struct HashImpl {
size_t operator()(const T& value) const { return CityHashState::hash(value); }
};
template <typename T>
struct Hash
: absl::conditional_t<is_hashable<T>::value, HashImpl<T>, PoisonedHash> {};
template <typename H>
template <typename T, typename... Ts>
H HashStateBase<H>::combine(H state, const T& value, const Ts&... values) {
return H::combine(hash_internal::HashSelect::template Apply<T>::Invoke(
std::move(state), value),
values...);
}
// HashStateBase::combine_contiguous()
template <typename H>
template <typename T>
H HashStateBase<H>::combine_contiguous(H state, const T* data, size_t size) {
return hash_internal::hash_range_or_bytes(std::move(state), data, size);
}
// HashStateBase::PiecewiseCombiner::add_buffer()
template <typename H>
H PiecewiseCombiner::add_buffer(H state, const unsigned char* data,
size_t size) {
if (position_ + size < PiecewiseChunkSize()) {
// This partial chunk does not fill our existing buffer
memcpy(buf_ + position_, data, size);
position_ += size;
return state;
}
// If the buffer is partially filled we need to complete the buffer
// and hash it.
if (position_ != 0) {
const size_t bytes_needed = PiecewiseChunkSize() - position_;
memcpy(buf_ + position_, data, bytes_needed);
state = H::combine_contiguous(std::move(state), buf_, PiecewiseChunkSize());
data += bytes_needed;
size -= bytes_needed;
}
// Hash whatever chunks we can without copying
while (size >= PiecewiseChunkSize()) {
state = H::combine_contiguous(std::move(state), data, PiecewiseChunkSize());
data += PiecewiseChunkSize();
size -= PiecewiseChunkSize();
}
// Fill the buffer with the remainder
memcpy(buf_, data, size);
position_ = size;
return state;
}
// HashStateBase::PiecewiseCombiner::finalize()
template <typename H>
H PiecewiseCombiner::finalize(H state) {
// Hash the remainder left in the buffer, which may be empty
return H::combine_contiguous(std::move(state), buf_, position_);
}
} // namespace hash_internal
ABSL_NAMESPACE_END
} // namespace absl
#endif // ABSL_HASH_INTERNAL_HASH_H_
|