• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2018 The Abseil Authors.
2 //
3 // Licensed under the Apache License, Version 2.0 (the "License");
4 // you may not use this file except in compliance with the License.
5 // You may obtain a copy of the License at
6 //
7 //      https://www.apache.org/licenses/LICENSE-2.0
8 //
9 // Unless required by applicable law or agreed to in writing, software
10 // distributed under the License is distributed on an "AS IS" BASIS,
11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 // See the License for the specific language governing permissions and
13 // limitations under the License.
14 //
15 // -----------------------------------------------------------------------------
16 // File: hash.h
17 // -----------------------------------------------------------------------------
18 //
19 #ifndef ABSL_HASH_INTERNAL_HASH_H_
20 #define ABSL_HASH_INTERNAL_HASH_H_
21 
22 #ifdef __APPLE__
23 #include <Availability.h>
24 #include <TargetConditionals.h>
25 #endif
26 
27 #include "absl/base/config.h"
28 
29 // For feature testing and determining which headers can be included.
30 #if ABSL_INTERNAL_CPLUSPLUS_LANG >= 202002L
31 #include <version>
32 #else
33 #include <ciso646>
34 #endif
35 
36 #include <algorithm>
37 #include <array>
38 #include <bitset>
39 #include <cassert>
40 #include <cmath>
41 #include <cstddef>
42 #include <cstdint>
43 #include <cstring>
44 #include <deque>
45 #include <forward_list>
46 #include <functional>
47 #include <iterator>
48 #include <limits>
49 #include <list>
50 #include <map>
51 #include <memory>
52 #include <set>
53 #include <string>
54 #include <tuple>
55 #include <type_traits>
56 #include <unordered_map>
57 #include <unordered_set>
58 #include <utility>
59 #include <vector>
60 
61 #include "absl/base/attributes.h"
62 #include "absl/base/internal/endian.h"
63 #include "absl/base/internal/unaligned_access.h"
64 #include "absl/base/optimization.h"
65 #include "absl/base/port.h"
66 #include "absl/container/fixed_array.h"
67 #include "absl/hash/internal/city.h"
68 #include "absl/hash/internal/low_level_hash.h"
69 #include "absl/meta/type_traits.h"
70 #include "absl/numeric/bits.h"
71 #include "absl/numeric/int128.h"
72 #include "absl/strings/string_view.h"
73 #include "absl/types/optional.h"
74 #include "absl/types/variant.h"
75 #include "absl/utility/utility.h"
76 
77 #if defined(__cpp_lib_filesystem) && __cpp_lib_filesystem >= 201703L
78 #include <filesystem>  // NOLINT
79 #endif
80 
81 #ifdef ABSL_HAVE_STD_STRING_VIEW
82 #include <string_view>
83 #endif
84 
85 #ifdef __ARM_ACLE
86 #include <arm_acle.h>
87 #endif
88 
89 namespace absl {
90 ABSL_NAMESPACE_BEGIN
91 
92 class HashState;
93 
94 namespace hash_internal {
95 
96 // Internal detail: Large buffers are hashed in smaller chunks.  This function
97 // returns the size of these chunks.
PiecewiseChunkSize()98 constexpr size_t PiecewiseChunkSize() { return 1024; }
99 
100 // PiecewiseCombiner
101 //
102 // PiecewiseCombiner is an internal-only helper class for hashing a piecewise
103 // buffer of `char` or `unsigned char` as though it were contiguous.  This class
104 // provides two methods:
105 //
106 //   H add_buffer(state, data, size)
107 //   H finalize(state)
108 //
109 // `add_buffer` can be called zero or more times, followed by a single call to
110 // `finalize`.  This will produce the same hash expansion as concatenating each
111 // buffer piece into a single contiguous buffer, and passing this to
112 // `H::combine_contiguous`.
113 //
114 //  Example usage:
115 //    PiecewiseCombiner combiner;
116 //    for (const auto& piece : pieces) {
117 //      state = combiner.add_buffer(std::move(state), piece.data, piece.size);
118 //    }
119 //    return combiner.finalize(std::move(state));
120 class PiecewiseCombiner {
121  public:
PiecewiseCombiner()122   PiecewiseCombiner() : position_(0) {}
123   PiecewiseCombiner(const PiecewiseCombiner&) = delete;
124   PiecewiseCombiner& operator=(const PiecewiseCombiner&) = delete;
125 
126   // PiecewiseCombiner::add_buffer()
127   //
128   // Appends the given range of bytes to the sequence to be hashed, which may
129   // modify the provided hash state.
130   template <typename H>
131   H add_buffer(H state, const unsigned char* data, size_t size);
132   template <typename H>
add_buffer(H state,const char * data,size_t size)133   H add_buffer(H state, const char* data, size_t size) {
134     return add_buffer(std::move(state),
135                       reinterpret_cast<const unsigned char*>(data), size);
136   }
137 
138   // PiecewiseCombiner::finalize()
139   //
140   // Finishes combining the hash sequence, which may may modify the provided
141   // hash state.
142   //
143   // Once finalize() is called, add_buffer() may no longer be called. The
144   // resulting hash state will be the same as if the pieces passed to
145   // add_buffer() were concatenated into a single flat buffer, and then provided
146   // to H::combine_contiguous().
147   template <typename H>
148   H finalize(H state);
149 
150  private:
151   unsigned char buf_[PiecewiseChunkSize()];
152   size_t position_;
153 };
154 
155 // is_hashable()
156 //
157 // Trait class which returns true if T is hashable by the absl::Hash framework.
158 // Used for the AbslHashValue implementations for composite types below.
159 template <typename T>
160 struct is_hashable;
161 
162 // HashStateBase
163 //
164 // An internal implementation detail that contains common implementation details
165 // for all of the "hash state objects" objects generated by Abseil.  This is not
166 // a public API; users should not create classes that inherit from this.
167 //
168 // A hash state object is the template argument `H` passed to `AbslHashValue`.
169 // It represents an intermediate state in the computation of an unspecified hash
170 // algorithm. `HashStateBase` provides a CRTP style base class for hash state
171 // implementations. Developers adding type support for `absl::Hash` should not
172 // rely on any parts of the state object other than the following member
173 // functions:
174 //
175 //   * HashStateBase::combine()
176 //   * HashStateBase::combine_contiguous()
177 //   * HashStateBase::combine_unordered()
178 //
179 // A derived hash state class of type `H` must provide a public member function
180 // with a signature similar to the following:
181 //
182 //    `static H combine_contiguous(H state, const unsigned char*, size_t)`.
183 //
184 // It must also provide a private template method named RunCombineUnordered.
185 //
186 // A "consumer" is a 1-arg functor returning void.  Its argument is a reference
187 // to an inner hash state object, and it may be called multiple times.  When
188 // called, the functor consumes the entropy from the provided state object,
189 // and resets that object to its empty state.
190 //
191 // A "combiner" is a stateless 2-arg functor returning void.  Its arguments are
192 // an inner hash state object and an ElementStateConsumer functor.  A combiner
193 // uses the provided inner hash state object to hash each element of the
194 // container, passing the inner hash state object to the consumer after hashing
195 // each element.
196 //
197 // Given these definitions, a derived hash state class of type H
198 // must provide a private template method with a signature similar to the
199 // following:
200 //
201 //    `template <typename CombinerT>`
202 //    `static H RunCombineUnordered(H outer_state, CombinerT combiner)`
203 //
204 // This function is responsible for constructing the inner state object and
205 // providing a consumer to the combiner.  It uses side effects of the consumer
206 // and combiner to mix the state of each element in an order-independent manner,
207 // and uses this to return an updated value of `outer_state`.
208 //
209 // This inside-out approach generates efficient object code in the normal case,
210 // but allows us to use stack storage to implement the absl::HashState type
211 // erasure mechanism (avoiding heap allocations while hashing).
212 //
213 // `HashStateBase` will provide a complete implementation for a hash state
214 // object in terms of these two methods.
215 //
216 // Example:
217 //
218 //   // Use CRTP to define your derived class.
219 //   struct MyHashState : HashStateBase<MyHashState> {
220 //       static H combine_contiguous(H state, const unsigned char*, size_t);
221 //       using MyHashState::HashStateBase::combine;
222 //       using MyHashState::HashStateBase::combine_contiguous;
223 //       using MyHashState::HashStateBase::combine_unordered;
224 //     private:
225 //       template <typename CombinerT>
226 //       static H RunCombineUnordered(H state, CombinerT combiner);
227 //   };
228 template <typename H>
229 class HashStateBase {
230  public:
231   // HashStateBase::combine()
232   //
233   // Combines an arbitrary number of values into a hash state, returning the
234   // updated state.
235   //
236   // Each of the value types `T` must be separately hashable by the Abseil
237   // hashing framework.
238   //
239   // NOTE:
240   //
241   //   state = H::combine(std::move(state), value1, value2, value3);
242   //
243   // is guaranteed to produce the same hash expansion as:
244   //
245   //   state = H::combine(std::move(state), value1);
246   //   state = H::combine(std::move(state), value2);
247   //   state = H::combine(std::move(state), value3);
248   template <typename T, typename... Ts>
249   static H combine(H state, const T& value, const Ts&... values);
combine(H state)250   static H combine(H state) { return state; }
251 
252   // HashStateBase::combine_contiguous()
253   //
254   // Combines a contiguous array of `size` elements into a hash state, returning
255   // the updated state.
256   //
257   // NOTE:
258   //
259   //   state = H::combine_contiguous(std::move(state), data, size);
260   //
261   // is NOT guaranteed to produce the same hash expansion as a for-loop (it may
262   // perform internal optimizations).  If you need this guarantee, use the
263   // for-loop instead.
264   template <typename T>
265   static H combine_contiguous(H state, const T* data, size_t size);
266 
267   template <typename I>
268   static H combine_unordered(H state, I begin, I end);
269 
270   using AbslInternalPiecewiseCombiner = PiecewiseCombiner;
271 
272   template <typename T>
273   using is_hashable = absl::hash_internal::is_hashable<T>;
274 
275  private:
276   // Common implementation of the iteration step of a "combiner", as described
277   // above.
278   template <typename I>
279   struct CombineUnorderedCallback {
280     I begin;
281     I end;
282 
283     template <typename InnerH, typename ElementStateConsumer>
operatorCombineUnorderedCallback284     void operator()(InnerH inner_state, ElementStateConsumer cb) {
285       for (; begin != end; ++begin) {
286         inner_state = H::combine(std::move(inner_state), *begin);
287         cb(inner_state);
288       }
289     }
290   };
291 };
292 
293 // is_uniquely_represented
294 //
295 // `is_uniquely_represented<T>` is a trait class that indicates whether `T`
296 // is uniquely represented.
297 //
298 // A type is "uniquely represented" if two equal values of that type are
299 // guaranteed to have the same bytes in their underlying storage. In other
300 // words, if `a == b`, then `memcmp(&a, &b, sizeof(T))` is guaranteed to be
301 // zero. This property cannot be detected automatically, so this trait is false
302 // by default, but can be specialized by types that wish to assert that they are
303 // uniquely represented. This makes them eligible for certain optimizations.
304 //
305 // If you have any doubt whatsoever, do not specialize this template.
306 // The default is completely safe, and merely disables some optimizations
307 // that will not matter for most types. Specializing this template,
308 // on the other hand, can be very hazardous.
309 //
310 // To be uniquely represented, a type must not have multiple ways of
311 // representing the same value; for example, float and double are not
312 // uniquely represented, because they have distinct representations for
313 // +0 and -0. Furthermore, the type's byte representation must consist
314 // solely of user-controlled data, with no padding bits and no compiler-
315 // controlled data such as vptrs or sanitizer metadata. This is usually
316 // very difficult to guarantee, because in most cases the compiler can
317 // insert data and padding bits at its own discretion.
318 //
319 // If you specialize this template for a type `T`, you must do so in the file
320 // that defines that type (or in this file). If you define that specialization
321 // anywhere else, `is_uniquely_represented<T>` could have different meanings
322 // in different places.
323 //
324 // The Enable parameter is meaningless; it is provided as a convenience,
325 // to support certain SFINAE techniques when defining specializations.
326 template <typename T, typename Enable = void>
327 struct is_uniquely_represented : std::false_type {};
328 
329 // is_uniquely_represented<unsigned char>
330 //
331 // unsigned char is a synonym for "byte", so it is guaranteed to be
332 // uniquely represented.
333 template <>
334 struct is_uniquely_represented<unsigned char> : std::true_type {};
335 
336 // is_uniquely_represented for non-standard integral types
337 //
338 // Integral types other than bool should be uniquely represented on any
339 // platform that this will plausibly be ported to.
340 template <typename Integral>
341 struct is_uniquely_represented<
342     Integral, typename std::enable_if<std::is_integral<Integral>::value>::type>
343     : std::true_type {};
344 
345 // is_uniquely_represented<bool>
346 //
347 //
348 template <>
349 struct is_uniquely_represented<bool> : std::false_type {};
350 
351 #ifdef ABSL_HAVE_INTRINSIC_INT128
352 // Specialize the trait for GNU extension types.
353 template <>
354 struct is_uniquely_represented<__int128> : std::true_type {};
355 template <>
356 struct is_uniquely_represented<unsigned __int128> : std::true_type {};
357 #endif  // ABSL_HAVE_INTRINSIC_INT128
358 
359 template <typename T>
360 struct FitsIn64Bits : std::integral_constant<bool, sizeof(T) <= 8> {};
361 
362 struct CombineRaw {
363   template <typename H>
364   H operator()(H state, uint64_t value) const {
365     return H::combine_raw(std::move(state), value);
366   }
367 };
368 
369 // hash_bytes()
370 //
371 // Convenience function that combines `hash_state` with the byte representation
372 // of `value`.
373 template <typename H, typename T,
374           absl::enable_if_t<FitsIn64Bits<T>::value, int> = 0>
375 H hash_bytes(H hash_state, const T& value) {
376   const unsigned char* start = reinterpret_cast<const unsigned char*>(&value);
377   uint64_t v;
378   if (sizeof(T) == 1) {
379     v = *start;
380   } else if (sizeof(T) == 2) {
381     v = absl::base_internal::UnalignedLoad16(start);
382   } else if (sizeof(T) == 4) {
383     v = absl::base_internal::UnalignedLoad32(start);
384   } else {
385     assert(sizeof(T) == 8);
386     v = absl::base_internal::UnalignedLoad64(start);
387   }
388   return CombineRaw()(std::move(hash_state), v);
389 }
390 template <typename H, typename T,
391           absl::enable_if_t<!FitsIn64Bits<T>::value, int> = 0>
392 H hash_bytes(H hash_state, const T& value) {
393   const unsigned char* start = reinterpret_cast<const unsigned char*>(&value);
394   return H::combine_contiguous(std::move(hash_state), start, sizeof(value));
395 }
396 
397 // -----------------------------------------------------------------------------
398 // AbslHashValue for Basic Types
399 // -----------------------------------------------------------------------------
400 
401 // Note: Default `AbslHashValue` implementations live in `hash_internal`. This
402 // allows us to block lexical scope lookup when doing an unqualified call to
403 // `AbslHashValue` below. User-defined implementations of `AbslHashValue` can
404 // only be found via ADL.
405 
406 // AbslHashValue() for hashing bool values
407 //
408 // We use SFINAE to ensure that this overload only accepts bool, not types that
409 // are convertible to bool.
410 template <typename H, typename B>
411 typename std::enable_if<std::is_same<B, bool>::value, H>::type AbslHashValue(
412     H hash_state, B value) {
413   return H::combine(std::move(hash_state),
414                     static_cast<unsigned char>(value ? 1 : 0));
415 }
416 
417 // AbslHashValue() for hashing enum values
418 template <typename H, typename Enum>
419 typename std::enable_if<std::is_enum<Enum>::value, H>::type AbslHashValue(
420     H hash_state, Enum e) {
421   // In practice, we could almost certainly just invoke hash_bytes directly,
422   // but it's possible that a sanitizer might one day want to
423   // store data in the unused bits of an enum. To avoid that risk, we
424   // convert to the underlying type before hashing. Hopefully this will get
425   // optimized away; if not, we can reopen discussion with c-toolchain-team.
426   return H::combine(std::move(hash_state),
427                     static_cast<typename std::underlying_type<Enum>::type>(e));
428 }
429 // AbslHashValue() for hashing floating-point values
430 template <typename H, typename Float>
431 typename std::enable_if<std::is_same<Float, float>::value ||
432                             std::is_same<Float, double>::value,
433                         H>::type
434 AbslHashValue(H hash_state, Float value) {
435   return hash_internal::hash_bytes(std::move(hash_state),
436                                    value == 0 ? 0 : value);
437 }
438 
439 // Long double has the property that it might have extra unused bytes in it.
440 // For example, in x86 sizeof(long double)==16 but it only really uses 80-bits
441 // of it. This means we can't use hash_bytes on a long double and have to
442 // convert it to something else first.
443 template <typename H, typename LongDouble>
444 typename std::enable_if<std::is_same<LongDouble, long double>::value, H>::type
445 AbslHashValue(H hash_state, LongDouble value) {
446   const int category = std::fpclassify(value);
447   switch (category) {
448     case FP_INFINITE:
449       // Add the sign bit to differentiate between +Inf and -Inf
450       hash_state = H::combine(std::move(hash_state), std::signbit(value));
451       break;
452 
453     case FP_NAN:
454     case FP_ZERO:
455     default:
456       // Category is enough for these.
457       break;
458 
459     case FP_NORMAL:
460     case FP_SUBNORMAL:
461       // We can't convert `value` directly to double because this would have
462       // undefined behavior if the value is out of range.
463       // std::frexp gives us a value in the range (-1, -.5] or [.5, 1) that is
464       // guaranteed to be in range for `double`. The truncation is
465       // implementation defined, but that works as long as it is deterministic.
466       int exp;
467       auto mantissa = static_cast<double>(std::frexp(value, &exp));
468       hash_state = H::combine(std::move(hash_state), mantissa, exp);
469   }
470 
471   return H::combine(std::move(hash_state), category);
472 }
473 
474 // Without this overload, an array decays to a pointer and we hash that, which
475 // is not likely to be what the caller intended.
476 template <typename H, typename T, size_t N>
477 H AbslHashValue(H hash_state, T (&)[N]) {
478   static_assert(
479       sizeof(T) == -1,
480       "Hashing C arrays is not allowed. For string literals, wrap the literal "
481       "in absl::string_view(). To hash the array contents, use "
482       "absl::MakeSpan() or make the array an std::array. To hash the array "
483       "address, use &array[0].");
484   return hash_state;
485 }
486 
487 // AbslHashValue() for hashing pointers
488 template <typename H, typename T>
489 std::enable_if_t<std::is_pointer<T>::value, H> AbslHashValue(H hash_state,
490                                                              T ptr) {
491   auto v = reinterpret_cast<uintptr_t>(ptr);
492   // Due to alignment, pointers tend to have low bits as zero, and the next few
493   // bits follow a pattern since they are also multiples of some base value. The
494   // byte swap in WeakMix helps ensure we still have good entropy in low bits.
495   // Mix pointers twice to ensure we have good entropy in low bits.
496   return H::combine(std::move(hash_state), v, v);
497 }
498 
499 // AbslHashValue() for hashing nullptr_t
500 template <typename H>
501 H AbslHashValue(H hash_state, std::nullptr_t) {
502   return H::combine(std::move(hash_state), static_cast<void*>(nullptr));
503 }
504 
505 // AbslHashValue() for hashing pointers-to-member
506 template <typename H, typename T, typename C>
507 H AbslHashValue(H hash_state, T C::*ptr) {
508   auto salient_ptm_size = [](std::size_t n) -> std::size_t {
509 #if defined(_MSC_VER)
510     // Pointers-to-member-function on MSVC consist of one pointer plus 0, 1, 2,
511     // or 3 ints. In 64-bit mode, they are 8-byte aligned and thus can contain
512     // padding (namely when they have 1 or 3 ints). The value below is a lower
513     // bound on the number of salient, non-padding bytes that we use for
514     // hashing.
515     if (alignof(T C::*) == alignof(int)) {
516       // No padding when all subobjects have the same size as the total
517       // alignment. This happens in 32-bit mode.
518       return n;
519     } else {
520       // Padding for 1 int (size 16) or 3 ints (size 24).
521       // With 2 ints, the size is 16 with no padding, which we pessimize.
522       return n == 24 ? 20 : n == 16 ? 12 : n;
523     }
524 #else
525   // On other platforms, we assume that pointers-to-members do not have
526   // padding.
527 #ifdef __cpp_lib_has_unique_object_representations
528     static_assert(std::has_unique_object_representations<T C::*>::value);
529 #endif  // __cpp_lib_has_unique_object_representations
530     return n;
531 #endif
532   };
533   return H::combine_contiguous(std::move(hash_state),
534                                reinterpret_cast<unsigned char*>(&ptr),
535                                salient_ptm_size(sizeof ptr));
536 }
537 
538 // -----------------------------------------------------------------------------
539 // AbslHashValue for Composite Types
540 // -----------------------------------------------------------------------------
541 
542 // AbslHashValue() for hashing pairs
543 template <typename H, typename T1, typename T2>
544 typename std::enable_if<is_hashable<T1>::value && is_hashable<T2>::value,
545                         H>::type
546 AbslHashValue(H hash_state, const std::pair<T1, T2>& p) {
547   return H::combine(std::move(hash_state), p.first, p.second);
548 }
549 
550 // hash_tuple()
551 //
552 // Helper function for hashing a tuple. The third argument should
553 // be an index_sequence running from 0 to tuple_size<Tuple> - 1.
554 template <typename H, typename Tuple, size_t... Is>
555 H hash_tuple(H hash_state, const Tuple& t, absl::index_sequence<Is...>) {
556   return H::combine(std::move(hash_state), std::get<Is>(t)...);
557 }
558 
559 // AbslHashValue for hashing tuples
560 template <typename H, typename... Ts>
561 #if defined(_MSC_VER)
562 // This SFINAE gets MSVC confused under some conditions. Let's just disable it
563 // for now.
564 H
565 #else   // _MSC_VER
566 typename std::enable_if<absl::conjunction<is_hashable<Ts>...>::value, H>::type
567 #endif  // _MSC_VER
568 AbslHashValue(H hash_state, const std::tuple<Ts...>& t) {
569   return hash_internal::hash_tuple(std::move(hash_state), t,
570                                    absl::make_index_sequence<sizeof...(Ts)>());
571 }
572 
573 // -----------------------------------------------------------------------------
574 // AbslHashValue for Pointers
575 // -----------------------------------------------------------------------------
576 
577 // AbslHashValue for hashing unique_ptr
578 template <typename H, typename T, typename D>
579 H AbslHashValue(H hash_state, const std::unique_ptr<T, D>& ptr) {
580   return H::combine(std::move(hash_state), ptr.get());
581 }
582 
583 // AbslHashValue for hashing shared_ptr
584 template <typename H, typename T>
585 H AbslHashValue(H hash_state, const std::shared_ptr<T>& ptr) {
586   return H::combine(std::move(hash_state), ptr.get());
587 }
588 
589 // -----------------------------------------------------------------------------
590 // AbslHashValue for String-Like Types
591 // -----------------------------------------------------------------------------
592 
593 // AbslHashValue for hashing strings
594 //
595 // All the string-like types supported here provide the same hash expansion for
596 // the same character sequence. These types are:
597 //
598 //  - `absl::Cord`
599 //  - `std::string` (and std::basic_string<T, std::char_traits<T>, A> for
600 //      any allocator A and any T in {char, wchar_t, char16_t, char32_t})
601 //  - `absl::string_view`, `std::string_view`, `std::wstring_view`,
602 //    `std::u16string_view`, and `std::u32_string_view`.
603 //
604 // For simplicity, we currently support only strings built on `char`, `wchar_t`,
605 // `char16_t`, or `char32_t`. This support may be broadened, if necessary, but
606 // with some caution - this overload would misbehave in cases where the traits'
607 // `eq()` member isn't equivalent to `==` on the underlying character type.
608 template <typename H>
609 H AbslHashValue(H hash_state, absl::string_view str) {
610   return H::combine(
611       H::combine_contiguous(std::move(hash_state), str.data(), str.size()),
612       str.size());
613 }
614 
615 // Support std::wstring, std::u16string and std::u32string.
616 template <typename Char, typename Alloc, typename H,
617           typename = absl::enable_if_t<std::is_same<Char, wchar_t>::value ||
618                                        std::is_same<Char, char16_t>::value ||
619                                        std::is_same<Char, char32_t>::value>>
620 H AbslHashValue(
621     H hash_state,
622     const std::basic_string<Char, std::char_traits<Char>, Alloc>& str) {
623   return H::combine(
624       H::combine_contiguous(std::move(hash_state), str.data(), str.size()),
625       str.size());
626 }
627 
628 #ifdef ABSL_HAVE_STD_STRING_VIEW
629 
630 // Support std::wstring_view, std::u16string_view and std::u32string_view.
631 template <typename Char, typename H,
632           typename = absl::enable_if_t<std::is_same<Char, wchar_t>::value ||
633                                        std::is_same<Char, char16_t>::value ||
634                                        std::is_same<Char, char32_t>::value>>
635 H AbslHashValue(H hash_state, std::basic_string_view<Char> str) {
636   return H::combine(
637       H::combine_contiguous(std::move(hash_state), str.data(), str.size()),
638       str.size());
639 }
640 
641 #endif  // ABSL_HAVE_STD_STRING_VIEW
642 
643 #if defined(__cpp_lib_filesystem) && __cpp_lib_filesystem >= 201703L && \
644     (!defined(__ENVIRONMENT_IPHONE_OS_VERSION_MIN_REQUIRED__) ||        \
645      __ENVIRONMENT_IPHONE_OS_VERSION_MIN_REQUIRED__ >= 130000) &&       \
646     (!defined(__ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__) ||         \
647      __ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__ >= 101500)
648 
649 #define ABSL_INTERNAL_STD_FILESYSTEM_PATH_HASH_AVAILABLE 1
650 
651 // Support std::filesystem::path. The SFINAE is required because some string
652 // types are implicitly convertible to std::filesystem::path.
653 template <typename Path, typename H,
654           typename = absl::enable_if_t<
655               std::is_same_v<Path, std::filesystem::path>>>
656 H AbslHashValue(H hash_state, const Path& path) {
657   // This is implemented by deferring to the standard library to compute the
658   // hash.  The standard library requires that for two paths, `p1 == p2`, then
659   // `hash_value(p1) == hash_value(p2)`. `AbslHashValue` has the same
660   // requirement. Since `operator==` does platform specific matching, deferring
661   // to the standard library is the simplest approach.
662   return H::combine(std::move(hash_state), std::filesystem::hash_value(path));
663 }
664 
665 #endif  // ABSL_INTERNAL_STD_FILESYSTEM_PATH_HASH_AVAILABLE
666 
667 // -----------------------------------------------------------------------------
668 // AbslHashValue for Sequence Containers
669 // -----------------------------------------------------------------------------
670 
671 // AbslHashValue for hashing std::array
672 template <typename H, typename T, size_t N>
673 typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
674     H hash_state, const std::array<T, N>& array) {
675   return H::combine_contiguous(std::move(hash_state), array.data(),
676                                array.size());
677 }
678 
679 // AbslHashValue for hashing std::deque
680 template <typename H, typename T, typename Allocator>
681 typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
682     H hash_state, const std::deque<T, Allocator>& deque) {
683   // TODO(gromer): investigate a more efficient implementation taking
684   // advantage of the chunk structure.
685   for (const auto& t : deque) {
686     hash_state = H::combine(std::move(hash_state), t);
687   }
688   return H::combine(std::move(hash_state), deque.size());
689 }
690 
691 // AbslHashValue for hashing std::forward_list
692 template <typename H, typename T, typename Allocator>
693 typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
694     H hash_state, const std::forward_list<T, Allocator>& list) {
695   size_t size = 0;
696   for (const T& t : list) {
697     hash_state = H::combine(std::move(hash_state), t);
698     ++size;
699   }
700   return H::combine(std::move(hash_state), size);
701 }
702 
703 // AbslHashValue for hashing std::list
704 template <typename H, typename T, typename Allocator>
705 typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
706     H hash_state, const std::list<T, Allocator>& list) {
707   for (const auto& t : list) {
708     hash_state = H::combine(std::move(hash_state), t);
709   }
710   return H::combine(std::move(hash_state), list.size());
711 }
712 
713 // AbslHashValue for hashing std::vector
714 //
715 // Do not use this for vector<bool> on platforms that have a working
716 // implementation of std::hash. It does not have a .data(), and a fallback for
717 // std::hash<> is most likely faster.
718 template <typename H, typename T, typename Allocator>
719 typename std::enable_if<is_hashable<T>::value && !std::is_same<T, bool>::value,
720                         H>::type
721 AbslHashValue(H hash_state, const std::vector<T, Allocator>& vector) {
722   return H::combine(H::combine_contiguous(std::move(hash_state), vector.data(),
723                                           vector.size()),
724                     vector.size());
725 }
726 
727 // AbslHashValue special cases for hashing std::vector<bool>
728 
729 #if defined(ABSL_IS_BIG_ENDIAN) && \
730     (defined(__GLIBCXX__) || defined(__GLIBCPP__))
731 
732 // std::hash in libstdc++ does not work correctly with vector<bool> on Big
733 // Endian platforms therefore we need to implement a custom AbslHashValue for
734 // it. More details on the bug:
735 // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102531
736 template <typename H, typename T, typename Allocator>
737 typename std::enable_if<is_hashable<T>::value && std::is_same<T, bool>::value,
738                         H>::type
739 AbslHashValue(H hash_state, const std::vector<T, Allocator>& vector) {
740   typename H::AbslInternalPiecewiseCombiner combiner;
741   for (const auto& i : vector) {
742     unsigned char c = static_cast<unsigned char>(i);
743     hash_state = combiner.add_buffer(std::move(hash_state), &c, sizeof(c));
744   }
745   return H::combine(combiner.finalize(std::move(hash_state)), vector.size());
746 }
747 #else
748 // When not working around the libstdc++ bug above, we still have to contend
749 // with the fact that std::hash<vector<bool>> is often poor quality, hashing
750 // directly on the internal words and on no other state.  On these platforms,
751 // vector<bool>{1, 1} and vector<bool>{1, 1, 0} hash to the same value.
752 //
753 // Mixing in the size (as we do in our other vector<> implementations) on top
754 // of the library-provided hash implementation avoids this QOI issue.
755 template <typename H, typename T, typename Allocator>
756 typename std::enable_if<is_hashable<T>::value && std::is_same<T, bool>::value,
757                         H>::type
758 AbslHashValue(H hash_state, const std::vector<T, Allocator>& vector) {
759   return H::combine(std::move(hash_state),
760                     std::hash<std::vector<T, Allocator>>{}(vector),
761                     vector.size());
762 }
763 #endif
764 
765 // -----------------------------------------------------------------------------
766 // AbslHashValue for Ordered Associative Containers
767 // -----------------------------------------------------------------------------
768 
769 // AbslHashValue for hashing std::map
770 template <typename H, typename Key, typename T, typename Compare,
771           typename Allocator>
772 typename std::enable_if<is_hashable<Key>::value && is_hashable<T>::value,
773                         H>::type
774 AbslHashValue(H hash_state, const std::map<Key, T, Compare, Allocator>& map) {
775   for (const auto& t : map) {
776     hash_state = H::combine(std::move(hash_state), t);
777   }
778   return H::combine(std::move(hash_state), map.size());
779 }
780 
781 // AbslHashValue for hashing std::multimap
782 template <typename H, typename Key, typename T, typename Compare,
783           typename Allocator>
784 typename std::enable_if<is_hashable<Key>::value && is_hashable<T>::value,
785                         H>::type
786 AbslHashValue(H hash_state,
787               const std::multimap<Key, T, Compare, Allocator>& map) {
788   for (const auto& t : map) {
789     hash_state = H::combine(std::move(hash_state), t);
790   }
791   return H::combine(std::move(hash_state), map.size());
792 }
793 
794 // AbslHashValue for hashing std::set
795 template <typename H, typename Key, typename Compare, typename Allocator>
796 typename std::enable_if<is_hashable<Key>::value, H>::type AbslHashValue(
797     H hash_state, const std::set<Key, Compare, Allocator>& set) {
798   for (const auto& t : set) {
799     hash_state = H::combine(std::move(hash_state), t);
800   }
801   return H::combine(std::move(hash_state), set.size());
802 }
803 
804 // AbslHashValue for hashing std::multiset
805 template <typename H, typename Key, typename Compare, typename Allocator>
806 typename std::enable_if<is_hashable<Key>::value, H>::type AbslHashValue(
807     H hash_state, const std::multiset<Key, Compare, Allocator>& set) {
808   for (const auto& t : set) {
809     hash_state = H::combine(std::move(hash_state), t);
810   }
811   return H::combine(std::move(hash_state), set.size());
812 }
813 
814 // -----------------------------------------------------------------------------
815 // AbslHashValue for Unordered Associative Containers
816 // -----------------------------------------------------------------------------
817 
818 // AbslHashValue for hashing std::unordered_set
819 template <typename H, typename Key, typename Hash, typename KeyEqual,
820           typename Alloc>
821 typename std::enable_if<is_hashable<Key>::value, H>::type AbslHashValue(
822     H hash_state, const std::unordered_set<Key, Hash, KeyEqual, Alloc>& s) {
823   return H::combine(
824       H::combine_unordered(std::move(hash_state), s.begin(), s.end()),
825       s.size());
826 }
827 
828 // AbslHashValue for hashing std::unordered_multiset
829 template <typename H, typename Key, typename Hash, typename KeyEqual,
830           typename Alloc>
831 typename std::enable_if<is_hashable<Key>::value, H>::type AbslHashValue(
832     H hash_state,
833     const std::unordered_multiset<Key, Hash, KeyEqual, Alloc>& s) {
834   return H::combine(
835       H::combine_unordered(std::move(hash_state), s.begin(), s.end()),
836       s.size());
837 }
838 
839 // AbslHashValue for hashing std::unordered_set
840 template <typename H, typename Key, typename T, typename Hash,
841           typename KeyEqual, typename Alloc>
842 typename std::enable_if<is_hashable<Key>::value && is_hashable<T>::value,
843                         H>::type
844 AbslHashValue(H hash_state,
845               const std::unordered_map<Key, T, Hash, KeyEqual, Alloc>& s) {
846   return H::combine(
847       H::combine_unordered(std::move(hash_state), s.begin(), s.end()),
848       s.size());
849 }
850 
851 // AbslHashValue for hashing std::unordered_multiset
852 template <typename H, typename Key, typename T, typename Hash,
853           typename KeyEqual, typename Alloc>
854 typename std::enable_if<is_hashable<Key>::value && is_hashable<T>::value,
855                         H>::type
856 AbslHashValue(H hash_state,
857               const std::unordered_multimap<Key, T, Hash, KeyEqual, Alloc>& s) {
858   return H::combine(
859       H::combine_unordered(std::move(hash_state), s.begin(), s.end()),
860       s.size());
861 }
862 
863 // -----------------------------------------------------------------------------
864 // AbslHashValue for Wrapper Types
865 // -----------------------------------------------------------------------------
866 
867 // AbslHashValue for hashing std::reference_wrapper
868 template <typename H, typename T>
869 typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
870     H hash_state, std::reference_wrapper<T> opt) {
871   return H::combine(std::move(hash_state), opt.get());
872 }
873 
874 // AbslHashValue for hashing absl::optional
875 template <typename H, typename T>
876 typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue(
877     H hash_state, const absl::optional<T>& opt) {
878   if (opt) hash_state = H::combine(std::move(hash_state), *opt);
879   return H::combine(std::move(hash_state), opt.has_value());
880 }
881 
882 // VariantVisitor
883 template <typename H>
884 struct VariantVisitor {
885   H&& hash_state;
886   template <typename T>
887   H operator()(const T& t) const {
888     return H::combine(std::move(hash_state), t);
889   }
890 };
891 
892 // AbslHashValue for hashing absl::variant
893 template <typename H, typename... T>
894 typename std::enable_if<conjunction<is_hashable<T>...>::value, H>::type
895 AbslHashValue(H hash_state, const absl::variant<T...>& v) {
896   if (!v.valueless_by_exception()) {
897     hash_state = absl::visit(VariantVisitor<H>{std::move(hash_state)}, v);
898   }
899   return H::combine(std::move(hash_state), v.index());
900 }
901 
902 // -----------------------------------------------------------------------------
903 // AbslHashValue for Other Types
904 // -----------------------------------------------------------------------------
905 
906 // AbslHashValue for hashing std::bitset is not defined on Little Endian
907 // platforms, for the same reason as for vector<bool> (see std::vector above):
908 // It does not expose the raw bytes, and a fallback to std::hash<> is most
909 // likely faster.
910 
911 #if defined(ABSL_IS_BIG_ENDIAN) && \
912     (defined(__GLIBCXX__) || defined(__GLIBCPP__))
913 // AbslHashValue for hashing std::bitset
914 //
915 // std::hash in libstdc++ does not work correctly with std::bitset on Big Endian
916 // platforms therefore we need to implement a custom AbslHashValue for it. More
917 // details on the bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102531
918 template <typename H, size_t N>
919 H AbslHashValue(H hash_state, const std::bitset<N>& set) {
920   typename H::AbslInternalPiecewiseCombiner combiner;
921   for (size_t i = 0; i < N; i++) {
922     unsigned char c = static_cast<unsigned char>(set[i]);
923     hash_state = combiner.add_buffer(std::move(hash_state), &c, sizeof(c));
924   }
925   return H::combine(combiner.finalize(std::move(hash_state)), N);
926 }
927 #endif
928 
929 // -----------------------------------------------------------------------------
930 
931 // hash_range_or_bytes()
932 //
933 // Mixes all values in the range [data, data+size) into the hash state.
934 // This overload accepts only uniquely-represented types, and hashes them by
935 // hashing the entire range of bytes.
936 template <typename H, typename T>
937 typename std::enable_if<is_uniquely_represented<T>::value, H>::type
938 hash_range_or_bytes(H hash_state, const T* data, size_t size) {
939   const auto* bytes = reinterpret_cast<const unsigned char*>(data);
940   return H::combine_contiguous(std::move(hash_state), bytes, sizeof(T) * size);
941 }
942 
943 // hash_range_or_bytes()
944 template <typename H, typename T>
945 typename std::enable_if<!is_uniquely_represented<T>::value, H>::type
946 hash_range_or_bytes(H hash_state, const T* data, size_t size) {
947   for (const auto end = data + size; data < end; ++data) {
948     hash_state = H::combine(std::move(hash_state), *data);
949   }
950   return hash_state;
951 }
952 
953 #if defined(ABSL_INTERNAL_LEGACY_HASH_NAMESPACE) && \
954     ABSL_META_INTERNAL_STD_HASH_SFINAE_FRIENDLY_
955 #define ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ 1
956 #else
957 #define ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ 0
958 #endif
959 
960 // HashSelect
961 //
962 // Type trait to select the appropriate hash implementation to use.
963 // HashSelect::type<T> will give the proper hash implementation, to be invoked
964 // as:
965 //   HashSelect::type<T>::Invoke(state, value)
966 // Also, HashSelect::type<T>::value is a boolean equal to `true` if there is a
967 // valid `Invoke` function. Types that are not hashable will have a ::value of
968 // `false`.
969 struct HashSelect {
970  private:
971   struct State : HashStateBase<State> {
972     static State combine_contiguous(State hash_state, const unsigned char*,
973                                     size_t);
974     using State::HashStateBase::combine_contiguous;
975     static State combine_raw(State state, uint64_t value);
976   };
977 
978   struct UniquelyRepresentedProbe {
979     template <typename H, typename T>
980     static auto Invoke(H state, const T& value)
981         -> absl::enable_if_t<is_uniquely_represented<T>::value, H> {
982       return hash_internal::hash_bytes(std::move(state), value);
983     }
984   };
985 
986   struct HashValueProbe {
987     template <typename H, typename T>
988     static auto Invoke(H state, const T& value) -> absl::enable_if_t<
989         std::is_same<H,
990                      decltype(AbslHashValue(std::move(state), value))>::value,
991         H> {
992       return AbslHashValue(std::move(state), value);
993     }
994   };
995 
996   struct LegacyHashProbe {
997 #if ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_
998     template <typename H, typename T>
999     static auto Invoke(H state, const T& value) -> absl::enable_if_t<
1000         std::is_convertible<
1001             decltype(ABSL_INTERNAL_LEGACY_HASH_NAMESPACE::hash<T>()(value)),
1002             size_t>::value,
1003         H> {
1004       return hash_internal::hash_bytes(
1005           std::move(state),
1006           ABSL_INTERNAL_LEGACY_HASH_NAMESPACE::hash<T>{}(value));
1007     }
1008 #endif  // ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_
1009   };
1010 
1011   struct StdHashProbe {
1012     template <typename H, typename T>
1013     static auto Invoke(H state, const T& value)
1014         -> absl::enable_if_t<type_traits_internal::IsHashable<T>::value, H> {
1015       return hash_internal::hash_bytes(std::move(state), std::hash<T>{}(value));
1016     }
1017   };
1018 
1019   template <typename Hash, typename T>
1020   struct Probe : Hash {
1021    private:
1022     template <typename H, typename = decltype(H::Invoke(
1023                               std::declval<State>(), std::declval<const T&>()))>
1024     static std::true_type Test(int);
1025     template <typename U>
1026     static std::false_type Test(char);
1027 
1028    public:
1029     static constexpr bool value = decltype(Test<Hash>(0))::value;
1030   };
1031 
1032  public:
1033   // Probe each implementation in order.
1034   // disjunction provides short circuiting wrt instantiation.
1035   template <typename T>
1036   using Apply = absl::disjunction<         //
1037       Probe<UniquelyRepresentedProbe, T>,  //
1038       Probe<HashValueProbe, T>,            //
1039       Probe<LegacyHashProbe, T>,           //
1040       Probe<StdHashProbe, T>,              //
1041       std::false_type>;
1042 };
1043 
1044 template <typename T>
1045 struct is_hashable
1046     : std::integral_constant<bool, HashSelect::template Apply<T>::value> {};
1047 
1048 // MixingHashState
1049 class ABSL_DLL MixingHashState : public HashStateBase<MixingHashState> {
1050   // absl::uint128 is not an alias or a thin wrapper around the intrinsic.
1051   // We use the intrinsic when available to improve performance.
1052 #ifdef ABSL_HAVE_INTRINSIC_INT128
1053   using uint128 = __uint128_t;
1054 #else   // ABSL_HAVE_INTRINSIC_INT128
1055   using uint128 = absl::uint128;
1056 #endif  // ABSL_HAVE_INTRINSIC_INT128
1057 
1058   // Random data taken from the hexadecimal digits of Pi's fractional component.
1059   // https://en.wikipedia.org/wiki/Nothing-up-my-sleeve_number
1060   ABSL_CACHELINE_ALIGNED static constexpr uint64_t kStaticRandomData[] = {
1061       0x243f'6a88'85a3'08d3, 0x1319'8a2e'0370'7344, 0xa409'3822'299f'31d0,
1062       0x082e'fa98'ec4e'6c89, 0x4528'21e6'38d0'1377,
1063   };
1064 
1065   static constexpr uint64_t kMul =
1066   sizeof(size_t) == 4 ? uint64_t{0xcc9e2d51}
1067                       : uint64_t{0xdcb22ca68cb134ed};
1068 
1069   template <typename T>
1070   using IntegralFastPath =
1071       conjunction<std::is_integral<T>, is_uniquely_represented<T>,
1072                   FitsIn64Bits<T>>;
1073 
1074  public:
1075   // Move only
1076   MixingHashState(MixingHashState&&) = default;
1077   MixingHashState& operator=(MixingHashState&&) = default;
1078 
1079   // MixingHashState::combine_contiguous()
1080   //
1081   // Fundamental base case for hash recursion: mixes the given range of bytes
1082   // into the hash state.
1083   static MixingHashState combine_contiguous(MixingHashState hash_state,
1084                                             const unsigned char* first,
1085                                             size_t size) {
1086     return MixingHashState(
1087         CombineContiguousImpl(hash_state.state_, first, size,
1088                               std::integral_constant<int, sizeof(size_t)>{}));
1089   }
1090   using MixingHashState::HashStateBase::combine_contiguous;
1091 
1092   // MixingHashState::hash()
1093   //
1094   // For performance reasons in non-opt mode, we specialize this for
1095   // integral types.
1096   // Otherwise we would be instantiating and calling dozens of functions for
1097   // something that is just one multiplication and a couple xor's.
1098   // The result should be the same as running the whole algorithm, but faster.
1099   template <typename T, absl::enable_if_t<IntegralFastPath<T>::value, int> = 0>
1100   static size_t hash(T value) {
1101     return static_cast<size_t>(
1102         WeakMix(Seed() ^ static_cast<std::make_unsigned_t<T>>(value)));
1103   }
1104 
1105   // Overload of MixingHashState::hash()
1106   template <typename T, absl::enable_if_t<!IntegralFastPath<T>::value, int> = 0>
1107   static size_t hash(const T& value) {
1108     return static_cast<size_t>(combine(MixingHashState{}, value).state_);
1109   }
1110 
1111  private:
1112   // Invoked only once for a given argument; that plus the fact that this is
1113   // move-only ensures that there is only one non-moved-from object.
1114   MixingHashState() : state_(Seed()) {}
1115 
1116   friend class MixingHashState::HashStateBase;
1117 
1118   template <typename CombinerT>
1119   static MixingHashState RunCombineUnordered(MixingHashState state,
1120                                              CombinerT combiner) {
1121     uint64_t unordered_state = 0;
1122     combiner(MixingHashState{}, [&](MixingHashState& inner_state) {
1123       // Add the hash state of the element to the running total, but mix the
1124       // carry bit back into the low bit.  This in intended to avoid losing
1125       // entropy to overflow, especially when unordered_multisets contain
1126       // multiple copies of the same value.
1127       auto element_state = inner_state.state_;
1128       unordered_state += element_state;
1129       if (unordered_state < element_state) {
1130         ++unordered_state;
1131       }
1132       inner_state = MixingHashState{};
1133     });
1134     return MixingHashState::combine(std::move(state), unordered_state);
1135   }
1136 
1137   // Allow the HashState type-erasure implementation to invoke
1138   // RunCombinedUnordered() directly.
1139   friend class absl::HashState;
1140   friend struct CombineRaw;
1141 
1142   // Workaround for MSVC bug.
1143   // We make the type copyable to fix the calling convention, even though we
1144   // never actually copy it. Keep it private to not affect the public API of the
1145   // type.
1146   MixingHashState(const MixingHashState&) = default;
1147 
1148   explicit MixingHashState(uint64_t state) : state_(state) {}
1149 
1150   // Combines a raw value from e.g. integrals/floats/pointers/etc. This allows
1151   // us to be consistent with IntegralFastPath when combining raw types, but
1152   // optimize Read1To3 and Read4To8 differently for the string case.
1153   static MixingHashState combine_raw(MixingHashState hash_state,
1154                                      uint64_t value) {
1155     return MixingHashState(WeakMix(hash_state.state_ ^ value));
1156   }
1157 
1158   // Implementation of the base case for combine_contiguous where we actually
1159   // mix the bytes into the state.
1160   // Dispatch to different implementations of the combine_contiguous depending
1161   // on the value of `sizeof(size_t)`.
1162   static uint64_t CombineContiguousImpl(uint64_t state,
1163                                         const unsigned char* first, size_t len,
1164                                         std::integral_constant<int, 4>
1165                                         /* sizeof_size_t */);
1166   static uint64_t CombineContiguousImpl(uint64_t state,
1167                                         const unsigned char* first, size_t len,
1168                                         std::integral_constant<int, 8>
1169                                         /* sizeof_size_t */);
1170 
1171   ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t CombineSmallContiguousImpl(
1172       uint64_t state, const unsigned char* first, size_t len) {
1173     ABSL_ASSUME(len <= 8);
1174     uint64_t v;
1175     if (len >= 4) {
1176       v = Read4To8(first, len);
1177     } else if (len > 0) {
1178       v = Read1To3(first, len);
1179     } else {
1180       // Empty ranges have no effect.
1181       return state;
1182     }
1183     return WeakMix(state ^ v);
1184   }
1185 
1186   ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t CombineContiguousImpl9to16(
1187       uint64_t state, const unsigned char* first, size_t len) {
1188     ABSL_ASSUME(len >= 9);
1189     ABSL_ASSUME(len <= 16);
1190     // Note: any time one half of the mix function becomes zero it will fail to
1191     // incorporate any bits from the other half. However, there is exactly 1 in
1192     // 2^64 values for each side that achieve this, and only when the size is
1193     // exactly 16 -- for smaller sizes there is an overlapping byte that makes
1194     // this impossible unless the seed is *also* incredibly unlucky.
1195     auto p = Read9To16(first, len);
1196     return Mix(state ^ p.first, kMul ^ p.second);
1197   }
1198 
1199   ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t CombineContiguousImpl17to32(
1200       uint64_t state, const unsigned char* first, size_t len) {
1201     ABSL_ASSUME(len >= 17);
1202     ABSL_ASSUME(len <= 32);
1203     // Do two mixes of overlapping 16-byte ranges in parallel to minimize
1204     // latency.
1205     const uint64_t m0 =
1206         Mix(Read8(first) ^ kStaticRandomData[1], Read8(first + 8) ^ state);
1207 
1208     const unsigned char* tail_16b_ptr = first + (len - 16);
1209     const uint64_t m1 = Mix(Read8(tail_16b_ptr) ^ kStaticRandomData[3],
1210                             Read8(tail_16b_ptr + 8) ^ state);
1211     return m0 ^ m1;
1212   }
1213 
1214   // Slow dispatch path for calls to CombineContiguousImpl with a size argument
1215   // larger than PiecewiseChunkSize().  Has the same effect as calling
1216   // CombineContiguousImpl() repeatedly with the chunk stride size.
1217   static uint64_t CombineLargeContiguousImpl32(uint64_t state,
1218                                                const unsigned char* first,
1219                                                size_t len);
1220   static uint64_t CombineLargeContiguousImpl64(uint64_t state,
1221                                                const unsigned char* first,
1222                                                size_t len);
1223 
1224   // Reads 9 to 16 bytes from p.
1225   // The least significant 8 bytes are in .first, the rest (zero padded) bytes
1226   // are in .second.
1227   static std::pair<uint64_t, uint64_t> Read9To16(const unsigned char* p,
1228                                                  size_t len) {
1229     uint64_t low_mem = Read8(p);
1230     uint64_t high_mem = Read8(p + len - 8);
1231 #ifdef ABSL_IS_LITTLE_ENDIAN
1232     uint64_t most_significant = high_mem;
1233     uint64_t least_significant = low_mem;
1234 #else
1235     uint64_t most_significant = low_mem;
1236     uint64_t least_significant = high_mem;
1237 #endif
1238     return {least_significant, most_significant};
1239   }
1240 
1241   // Reads 8 bytes from p.
1242   static uint64_t Read8(const unsigned char* p) {
1243     // Suppress erroneous array bounds errors on GCC.
1244 #if defined(__GNUC__) && !defined(__clang__)
1245 #pragma GCC diagnostic push
1246 #pragma GCC diagnostic ignored "-Warray-bounds"
1247 #endif
1248     return absl::base_internal::UnalignedLoad64(p);
1249 #if defined(__GNUC__) && !defined(__clang__)
1250 #pragma GCC diagnostic pop
1251 #endif
1252   }
1253 
1254   // Reads 4 to 8 bytes from p. Zero pads to fill uint64_t.
1255   // TODO(b/384509507): consider optimizing this by not requiring the output to
1256   // be equivalent to an integer load for 4/8 bytes. Currently, we rely on this
1257   // behavior for the HashConsistentAcrossIntTypes test case. Ditto for
1258   // Read1To3.
1259   static uint64_t Read4To8(const unsigned char* p, size_t len) {
1260     // If `len < 8`, we duplicate bytes in the middle.
1261     // E.g.:
1262     // `ABCD` will be read as `ABCDABCD`.
1263     // `ABCDE` will be read as `ABCDBCDE`.
1264     // `ABCDEF` will be read as `ABCDCDEF`.
1265     // `ABCDEFG` will be read as `ABCDDEFG`.
1266     // We also do not care about endianness. On big-endian platforms, bytes will
1267     // be shuffled (it's fine). We always shift low memory by 32, because that
1268     // can be pipelined earlier. Reading high memory requires computing
1269     // `p + len - 4`.
1270     uint64_t most_significant =
1271         static_cast<uint64_t>(absl::base_internal::UnalignedLoad32(p)) << 32;
1272     uint64_t least_significant =
1273         absl::base_internal::UnalignedLoad32(p + len - 4);
1274     return most_significant | least_significant;
1275   }
1276 
1277   // Reads 1 to 3 bytes from p. Zero pads to fill uint32_t.
1278   static uint32_t Read1To3(const unsigned char* p, size_t len) {
1279     // The trick used by this implementation is to avoid branches.
1280     // We always read three bytes by duplicating.
1281     // E.g.,
1282     // `A` is read as `AAA`.
1283     // `AB` is read as `ABB`.
1284     // `ABC` is read as `ABC`.
1285     // We always shift `p[0]` so that it can be pipelined better.
1286     // Other bytes require extra computation to find indices.
1287     uint32_t mem0 = (static_cast<uint32_t>(p[0]) << 16) | p[len - 1];
1288     uint32_t mem1 = static_cast<uint32_t>(p[len / 2]) << 8;
1289     return mem0 | mem1;
1290   }
1291 
1292   ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t Mix(uint64_t lhs, uint64_t rhs) {
1293     // Though the 128-bit product on AArch64 needs two instructions, it is
1294     // still a good balance between speed and hash quality.
1295     using MultType =
1296         absl::conditional_t<sizeof(size_t) == 4, uint64_t, uint128>;
1297     MultType m = lhs;
1298     m *= rhs;
1299     return static_cast<uint64_t>(m ^ (m >> (sizeof(m) * 8 / 2)));
1300   }
1301 
1302   // Slightly lower latency than Mix, but with lower quality. The byte swap
1303   // helps ensure that low bits still have high quality.
1304   ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t WeakMix(uint64_t n) {
1305     // WeakMix doesn't work well on 32-bit platforms so just use Mix.
1306     if (sizeof(size_t) < 8) return Mix(n, kMul);
1307 #ifdef __ARM_ACLE
1308     // gbswap_64 compiles to `rev` on ARM, but `rbit` is better because it
1309     // reverses bits rather than reversing bytes.
1310     return __rbitll(n * kMul);
1311 #else
1312     return absl::gbswap_64(n * kMul);
1313 #endif
1314   }
1315 
1316   // An extern to avoid bloat on a direct call to LowLevelHash() with fixed
1317   // values for both the seed and salt parameters.
1318   static uint64_t LowLevelHashImpl(const unsigned char* data, size_t len);
1319 
1320   ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t Hash64(const unsigned char* data,
1321                                                       size_t len) {
1322 #ifdef ABSL_HAVE_INTRINSIC_INT128
1323     return LowLevelHashImpl(data, len);
1324 #else
1325     return hash_internal::CityHash64(reinterpret_cast<const char*>(data), len);
1326 #endif
1327   }
1328 
1329   // Seed()
1330   //
1331   // A non-deterministic seed.
1332   //
1333   // The current purpose of this seed is to generate non-deterministic results
1334   // and prevent having users depend on the particular hash values.
1335   // It is not meant as a security feature right now, but it leaves the door
1336   // open to upgrade it to a true per-process random seed. A true random seed
1337   // costs more and we don't need to pay for that right now.
1338   //
1339   // On platforms with ASLR, we take advantage of it to make a per-process
1340   // random value.
1341   // See https://en.wikipedia.org/wiki/Address_space_layout_randomization
1342   //
1343   // On other platforms this is still going to be non-deterministic but most
1344   // probably per-build and not per-process.
1345   ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t Seed() {
1346 #if (!defined(__clang__) || __clang_major__ > 11) && \
1347     (!defined(__apple_build_version__) ||            \
1348      __apple_build_version__ >= 19558921)  // Xcode 12
1349     return static_cast<uint64_t>(reinterpret_cast<uintptr_t>(&kSeed));
1350 #else
1351     // Workaround the absence of
1352     // https://github.com/llvm/llvm-project/commit/bc15bf66dcca76cc06fe71fca35b74dc4d521021.
1353     return static_cast<uint64_t>(reinterpret_cast<uintptr_t>(kSeed));
1354 #endif
1355   }
1356   static const void* const kSeed;
1357 
1358   uint64_t state_;
1359 };
1360 
1361 // MixingHashState::CombineContiguousImpl()
1362 inline uint64_t MixingHashState::CombineContiguousImpl(
1363     uint64_t state, const unsigned char* first, size_t len,
1364     std::integral_constant<int, 4> /* sizeof_size_t */) {
1365   // For large values we use CityHash, for small ones we just use a
1366   // multiplicative hash.
1367   if (len <= 8) {
1368     return CombineSmallContiguousImpl(state, first, len);
1369   }
1370   if (ABSL_PREDICT_TRUE(len <= PiecewiseChunkSize())) {
1371     return Mix(state ^ hash_internal::CityHash32(
1372                            reinterpret_cast<const char*>(first), len),
1373                kMul);
1374   }
1375   return CombineLargeContiguousImpl32(state, first, len);
1376 }
1377 
1378 // Overload of MixingHashState::CombineContiguousImpl()
1379 inline uint64_t MixingHashState::CombineContiguousImpl(
1380     uint64_t state, const unsigned char* first, size_t len,
1381     std::integral_constant<int, 8> /* sizeof_size_t */) {
1382   // For large values we use LowLevelHash or CityHash depending on the platform,
1383   // for small ones we just use a multiplicative hash.
1384   if (len <= 8) {
1385     return CombineSmallContiguousImpl(state, first, len);
1386   }
1387   if (len <= 16) {
1388     return CombineContiguousImpl9to16(state, first, len);
1389   }
1390   if (len <= 32) {
1391     return CombineContiguousImpl17to32(state, first, len);
1392   }
1393   if (ABSL_PREDICT_TRUE(len <= PiecewiseChunkSize())) {
1394     return Mix(state ^ Hash64(first, len), kMul);
1395   }
1396   return CombineLargeContiguousImpl64(state, first, len);
1397 }
1398 
1399 struct AggregateBarrier {};
1400 
1401 // HashImpl
1402 
1403 // Add a private base class to make sure this type is not an aggregate.
1404 // Aggregates can be aggregate initialized even if the default constructor is
1405 // deleted.
1406 struct PoisonedHash : private AggregateBarrier {
1407   PoisonedHash() = delete;
1408   PoisonedHash(const PoisonedHash&) = delete;
1409   PoisonedHash& operator=(const PoisonedHash&) = delete;
1410 };
1411 
1412 template <typename T>
1413 struct HashImpl {
1414   size_t operator()(const T& value) const {
1415     return MixingHashState::hash(value);
1416   }
1417 };
1418 
1419 template <typename T>
1420 struct Hash
1421     : absl::conditional_t<is_hashable<T>::value, HashImpl<T>, PoisonedHash> {};
1422 
1423 template <typename H>
1424 template <typename T, typename... Ts>
1425 H HashStateBase<H>::combine(H state, const T& value, const Ts&... values) {
1426   return H::combine(hash_internal::HashSelect::template Apply<T>::Invoke(
1427                         std::move(state), value),
1428                     values...);
1429 }
1430 
1431 // HashStateBase::combine_contiguous()
1432 template <typename H>
1433 template <typename T>
1434 H HashStateBase<H>::combine_contiguous(H state, const T* data, size_t size) {
1435   return hash_internal::hash_range_or_bytes(std::move(state), data, size);
1436 }
1437 
1438 // HashStateBase::combine_unordered()
1439 template <typename H>
1440 template <typename I>
1441 H HashStateBase<H>::combine_unordered(H state, I begin, I end) {
1442   return H::RunCombineUnordered(std::move(state),
1443                                 CombineUnorderedCallback<I>{begin, end});
1444 }
1445 
1446 // HashStateBase::PiecewiseCombiner::add_buffer()
1447 template <typename H>
1448 H PiecewiseCombiner::add_buffer(H state, const unsigned char* data,
1449                                 size_t size) {
1450   if (position_ + size < PiecewiseChunkSize()) {
1451     // This partial chunk does not fill our existing buffer
1452     memcpy(buf_ + position_, data, size);
1453     position_ += size;
1454     return state;
1455   }
1456 
1457   // If the buffer is partially filled we need to complete the buffer
1458   // and hash it.
1459   if (position_ != 0) {
1460     const size_t bytes_needed = PiecewiseChunkSize() - position_;
1461     memcpy(buf_ + position_, data, bytes_needed);
1462     state = H::combine_contiguous(std::move(state), buf_, PiecewiseChunkSize());
1463     data += bytes_needed;
1464     size -= bytes_needed;
1465   }
1466 
1467   // Hash whatever chunks we can without copying
1468   while (size >= PiecewiseChunkSize()) {
1469     state = H::combine_contiguous(std::move(state), data, PiecewiseChunkSize());
1470     data += PiecewiseChunkSize();
1471     size -= PiecewiseChunkSize();
1472   }
1473   // Fill the buffer with the remainder
1474   memcpy(buf_, data, size);
1475   position_ = size;
1476   return state;
1477 }
1478 
1479 // HashStateBase::PiecewiseCombiner::finalize()
1480 template <typename H>
1481 H PiecewiseCombiner::finalize(H state) {
1482   // Hash the remainder left in the buffer, which may be empty
1483   return H::combine_contiguous(std::move(state), buf_, position_);
1484 }
1485 
1486 }  // namespace hash_internal
1487 ABSL_NAMESPACE_END
1488 }  // namespace absl
1489 
1490 #endif  // ABSL_HASH_INTERNAL_HASH_H_
1491