michael@0: // Copyright (c) 2011 The Chromium Authors. All rights reserved.
michael@0: // Use of this source code is governed by a BSD-style license that can be
michael@0: // found in the LICENSE file.
michael@0: //
michael@0: 
michael@0: //
michael@0: // Deal with the differences between Microsoft and GNU implemenations
michael@0: // of hash_map. Allows all platforms to use |base::hash_map| and
michael@0: // |base::hash_set|.
michael@0: //  eg:
michael@0: //   base::hash_map<int> my_map;
michael@0: //   base::hash_set<int> my_set;
michael@0: //
michael@0: // NOTE: It is an explicit non-goal of this class to provide a generic hash
michael@0: // function for pointers.  If you want to hash a pointers to a particular class,
michael@0: // please define the template specialization elsewhere (for example, in its
michael@0: // header file) and keep it specific to just pointers to that class.  This is
michael@0: // because identity hashes are not desirable for all types that might show up
michael@0: // in containers as pointers.
michael@0: 
michael@0: #ifndef BASE_CONTAINERS_HASH_TABLES_H_
michael@0: #define BASE_CONTAINERS_HASH_TABLES_H_
michael@0: 
michael@0: #include <utility>
michael@0: 
michael@0: #include "base/basictypes.h"
michael@0: #include "base/strings/string16.h"
michael@0: #include "build/build_config.h"
michael@0: 
michael@0: #if defined(COMPILER_MSVC)
michael@0: #include <hash_map>
michael@0: #include <hash_set>
michael@0: 
michael@0: #define BASE_HASH_NAMESPACE stdext
michael@0: 
michael@0: #elif defined(COMPILER_GCC)
michael@0: #if defined(OS_ANDROID)
michael@0: #define BASE_HASH_NAMESPACE std
michael@0: #else
michael@0: #define BASE_HASH_NAMESPACE __gnu_cxx
michael@0: #endif
michael@0: 
michael@0: // This is a hack to disable the gcc 4.4 warning about hash_map and hash_set
michael@0: // being deprecated.  We can get rid of this when we upgrade to VS2008 and we
michael@0: // can use <tr1/unordered_map> and <tr1/unordered_set>.
michael@0: #ifdef __DEPRECATED
michael@0: #define CHROME_OLD__DEPRECATED __DEPRECATED
michael@0: #undef __DEPRECATED
michael@0: #endif
michael@0: 
michael@0: #if defined(OS_ANDROID)
michael@0: #include <hash_map>
michael@0: #include <hash_set>
michael@0: #else
michael@0: #include <ext/hash_map>
michael@0: #include <ext/hash_set>
michael@0: #endif
michael@0: 
michael@0: #include <string>
michael@0: 
michael@0: #ifdef CHROME_OLD__DEPRECATED
michael@0: #define __DEPRECATED CHROME_OLD__DEPRECATED
michael@0: #undef CHROME_OLD__DEPRECATED
michael@0: #endif
michael@0: 
michael@0: namespace BASE_HASH_NAMESPACE {
michael@0: 
michael@0: #if !defined(OS_ANDROID)
michael@0: // The GNU C++ library provides identity hash functions for many integral types,
michael@0: // but not for |long long|.  This hash function will truncate if |size_t| is
michael@0: // narrower than |long long|.  This is probably good enough for what we will
michael@0: // use it for.
michael@0: 
michael@0: #define DEFINE_TRIVIAL_HASH(integral_type) \
michael@0:     template<> \
michael@0:     struct hash<integral_type> { \
michael@0:       std::size_t operator()(integral_type value) const { \
michael@0:         return static_cast<std::size_t>(value); \
michael@0:       } \
michael@0:     }
michael@0: 
michael@0: DEFINE_TRIVIAL_HASH(long long);
michael@0: DEFINE_TRIVIAL_HASH(unsigned long long);
michael@0: 
michael@0: #undef DEFINE_TRIVIAL_HASH
michael@0: #endif  // !defined(OS_ANDROID)
michael@0: 
michael@0: // Implement string hash functions so that strings of various flavors can
michael@0: // be used as keys in STL maps and sets.  The hash algorithm comes from the
michael@0: // GNU C++ library, in <tr1/functional>.  It is duplicated here because GCC
michael@0: // versions prior to 4.3.2 are unable to compile <tr1/functional> when RTTI
michael@0: // is disabled, as it is in our build.
michael@0: 
michael@0: #define DEFINE_STRING_HASH(string_type) \
michael@0:     template<> \
michael@0:     struct hash<string_type> { \
michael@0:       std::size_t operator()(const string_type& s) const { \
michael@0:         std::size_t result = 0; \
michael@0:         for (string_type::const_iterator i = s.begin(); i != s.end(); ++i) \
michael@0:           result = (result * 131) + *i; \
michael@0:         return result; \
michael@0:       } \
michael@0:     }
michael@0: 
michael@0: DEFINE_STRING_HASH(std::string);
michael@0: DEFINE_STRING_HASH(string16);
michael@0: 
michael@0: #undef DEFINE_STRING_HASH
michael@0: 
michael@0: }  // namespace BASE_HASH_NAMESPACE
michael@0: 
michael@0: #else  // COMPILER
michael@0: #error define BASE_HASH_NAMESPACE for your compiler
michael@0: #endif  // COMPILER
michael@0: 
michael@0: namespace base {
michael@0: using BASE_HASH_NAMESPACE::hash_map;
michael@0: using BASE_HASH_NAMESPACE::hash_multimap;
michael@0: using BASE_HASH_NAMESPACE::hash_multiset;
michael@0: using BASE_HASH_NAMESPACE::hash_set;
michael@0: 
michael@0: // Implement hashing for pairs of at-most 32 bit integer values.
michael@0: // When size_t is 32 bits, we turn the 64-bit hash code into 32 bits by using
michael@0: // multiply-add hashing. This algorithm, as described in
michael@0: // Theorem 4.3.3 of the thesis "Über die Komplexität der Multiplikation in
michael@0: // eingeschränkten Branchingprogrammmodellen" by Woelfel, is:
michael@0: //
michael@0: //   h32(x32, y32) = (h64(x32, y32) * rand_odd64 + rand16 * 2^16) % 2^64 / 2^32
michael@0: //
michael@0: // Contact danakj@chromium.org for any questions.
michael@0: inline std::size_t HashInts32(uint32 value1, uint32 value2) {
michael@0:   uint64 value1_64 = value1;
michael@0:   uint64 hash64 = (value1_64 << 32) | value2;
michael@0: 
michael@0:   if (sizeof(std::size_t) >= sizeof(uint64))
michael@0:     return static_cast<std::size_t>(hash64);
michael@0: 
michael@0:   uint64 odd_random = 481046412LL << 32 | 1025306955LL;
michael@0:   uint32 shift_random = 10121U << 16;
michael@0: 
michael@0:   hash64 = hash64 * odd_random + shift_random;
michael@0:   std::size_t high_bits = static_cast<std::size_t>(
michael@0:       hash64 >> (sizeof(uint64) - sizeof(std::size_t)));
michael@0:   return high_bits;
michael@0: }
michael@0: 
michael@0: // Implement hashing for pairs of up-to 64-bit integer values.
michael@0: // We use the compound integer hash method to produce a 64-bit hash code, by
michael@0: // breaking the two 64-bit inputs into 4 32-bit values:
michael@0: // http://opendatastructures.org/versions/edition-0.1d/ods-java/node33.html#SECTION00832000000000000000
michael@0: // Then we reduce our result to 32 bits if required, similar to above.
michael@0: inline std::size_t HashInts64(uint64 value1, uint64 value2) {
michael@0:   uint32 short_random1 = 842304669U;
michael@0:   uint32 short_random2 = 619063811U;
michael@0:   uint32 short_random3 = 937041849U;
michael@0:   uint32 short_random4 = 3309708029U;
michael@0: 
michael@0:   uint32 value1a = static_cast<uint32>(value1 & 0xffffffff);
michael@0:   uint32 value1b = static_cast<uint32>((value1 >> 32) & 0xffffffff);
michael@0:   uint32 value2a = static_cast<uint32>(value2 & 0xffffffff);
michael@0:   uint32 value2b = static_cast<uint32>((value2 >> 32) & 0xffffffff);
michael@0: 
michael@0:   uint64 product1 = static_cast<uint64>(value1a) * short_random1;
michael@0:   uint64 product2 = static_cast<uint64>(value1b) * short_random2;
michael@0:   uint64 product3 = static_cast<uint64>(value2a) * short_random3;
michael@0:   uint64 product4 = static_cast<uint64>(value2b) * short_random4;
michael@0: 
michael@0:   uint64 hash64 = product1 + product2 + product3 + product4;
michael@0: 
michael@0:   if (sizeof(std::size_t) >= sizeof(uint64))
michael@0:     return static_cast<std::size_t>(hash64);
michael@0: 
michael@0:   uint64 odd_random = 1578233944LL << 32 | 194370989LL;
michael@0:   uint32 shift_random = 20591U << 16;
michael@0: 
michael@0:   hash64 = hash64 * odd_random + shift_random;
michael@0:   std::size_t high_bits = static_cast<std::size_t>(
michael@0:       hash64 >> (sizeof(uint64) - sizeof(std::size_t)));
michael@0:   return high_bits;
michael@0: }
michael@0: 
michael@0: #define DEFINE_32BIT_PAIR_HASH(Type1, Type2) \
michael@0: inline std::size_t HashPair(Type1 value1, Type2 value2) { \
michael@0:   return HashInts32(value1, value2); \
michael@0: }
michael@0: 
michael@0: DEFINE_32BIT_PAIR_HASH(int16, int16);
michael@0: DEFINE_32BIT_PAIR_HASH(int16, uint16);
michael@0: DEFINE_32BIT_PAIR_HASH(int16, int32);
michael@0: DEFINE_32BIT_PAIR_HASH(int16, uint32);
michael@0: DEFINE_32BIT_PAIR_HASH(uint16, int16);
michael@0: DEFINE_32BIT_PAIR_HASH(uint16, uint16);
michael@0: DEFINE_32BIT_PAIR_HASH(uint16, int32);
michael@0: DEFINE_32BIT_PAIR_HASH(uint16, uint32);
michael@0: DEFINE_32BIT_PAIR_HASH(int32, int16);
michael@0: DEFINE_32BIT_PAIR_HASH(int32, uint16);
michael@0: DEFINE_32BIT_PAIR_HASH(int32, int32);
michael@0: DEFINE_32BIT_PAIR_HASH(int32, uint32);
michael@0: DEFINE_32BIT_PAIR_HASH(uint32, int16);
michael@0: DEFINE_32BIT_PAIR_HASH(uint32, uint16);
michael@0: DEFINE_32BIT_PAIR_HASH(uint32, int32);
michael@0: DEFINE_32BIT_PAIR_HASH(uint32, uint32);
michael@0: 
michael@0: #undef DEFINE_32BIT_PAIR_HASH
michael@0: 
michael@0: #define DEFINE_64BIT_PAIR_HASH(Type1, Type2) \
michael@0: inline std::size_t HashPair(Type1 value1, Type2 value2) { \
michael@0:   return HashInts64(value1, value2); \
michael@0: }
michael@0: 
michael@0: DEFINE_64BIT_PAIR_HASH(int16, int64);
michael@0: DEFINE_64BIT_PAIR_HASH(int16, uint64);
michael@0: DEFINE_64BIT_PAIR_HASH(uint16, int64);
michael@0: DEFINE_64BIT_PAIR_HASH(uint16, uint64);
michael@0: DEFINE_64BIT_PAIR_HASH(int32, int64);
michael@0: DEFINE_64BIT_PAIR_HASH(int32, uint64);
michael@0: DEFINE_64BIT_PAIR_HASH(uint32, int64);
michael@0: DEFINE_64BIT_PAIR_HASH(uint32, uint64);
michael@0: DEFINE_64BIT_PAIR_HASH(int64, int16);
michael@0: DEFINE_64BIT_PAIR_HASH(int64, uint16);
michael@0: DEFINE_64BIT_PAIR_HASH(int64, int32);
michael@0: DEFINE_64BIT_PAIR_HASH(int64, uint32);
michael@0: DEFINE_64BIT_PAIR_HASH(int64, int64);
michael@0: DEFINE_64BIT_PAIR_HASH(int64, uint64);
michael@0: DEFINE_64BIT_PAIR_HASH(uint64, int16);
michael@0: DEFINE_64BIT_PAIR_HASH(uint64, uint16);
michael@0: DEFINE_64BIT_PAIR_HASH(uint64, int32);
michael@0: DEFINE_64BIT_PAIR_HASH(uint64, uint32);
michael@0: DEFINE_64BIT_PAIR_HASH(uint64, int64);
michael@0: DEFINE_64BIT_PAIR_HASH(uint64, uint64);
michael@0: 
michael@0: #undef DEFINE_64BIT_PAIR_HASH
michael@0: }  // namespace base
michael@0: 
michael@0: namespace BASE_HASH_NAMESPACE {
michael@0: 
michael@0: // Implement methods for hashing a pair of integers, so they can be used as
michael@0: // keys in STL containers.
michael@0: 
michael@0: #if defined(COMPILER_MSVC)
michael@0: 
michael@0: template<typename Type1, typename Type2>
michael@0: inline std::size_t hash_value(const std::pair<Type1, Type2>& value) {
michael@0:   return base::HashPair(value.first, value.second);
michael@0: }
michael@0: 
michael@0: #elif defined(COMPILER_GCC)
michael@0: template<typename Type1, typename Type2>
michael@0: struct hash<std::pair<Type1, Type2> > {
michael@0:   std::size_t operator()(std::pair<Type1, Type2> value) const {
michael@0:     return base::HashPair(value.first, value.second);
michael@0:   }
michael@0: };
michael@0: 
michael@0: #else
michael@0: #error define hash<std::pair<Type1, Type2> > for your compiler
michael@0: #endif  // COMPILER
michael@0: 
michael@0: }
michael@0: 
michael@0: #undef DEFINE_PAIR_HASH_FUNCTION_START
michael@0: #undef DEFINE_PAIR_HASH_FUNCTION_END
michael@0: 
michael@0: #endif  // BASE_CONTAINERS_HASH_TABLES_H_