2 * Copyright (C) 2005, 2006, 2008, 2010, 2013 Apple Inc. All rights reserved.
3 * Copyright (C) 2010 Patrick Gansterer <paroga@paroga.com>
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Library General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Library General Public License for more details.
15 * You should have received a copy of the GNU Library General Public License
16 * along with this library; see the file COPYING.LIB. If not, write to
17 * the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
18 * Boston, MA 02110-1301, USA.
22 #ifndef WTF_StringHasher_h
23 #define WTF_StringHasher_h
25 #include "wtf/unicode/Unicode.h"
29 // Paul Hsieh's SuperFastHash
30 // http://www.azillionmonkeys.com/qed/hash.html
32 // LChar data is interpreted as Latin-1-encoded (zero extended to 16 bits).
34 // NOTE: The hash computation here must stay in sync with the create_hash_table script in
35 // JavaScriptCore and the CodeGeneratorJS.pm script in WebCore.
37 // Golden ratio. Arbitrary start value to avoid mapping all zeros to a hash value of zero.
38 static const unsigned stringHashingStartValue = 0x9E3779B9U;
42 static const unsigned flagCount = 8; // Save 8 bits for StringImpl to use as flags.
45 : m_hash(stringHashingStartValue)
46 , m_hasPendingCharacter(false)
47 , m_pendingCharacter(0)
51 // The hasher hashes two characters at a time, and thus an "aligned" hasher is one
52 // where an even number of characters have been added. Callers that always add
53 // characters two at a time can use the "assuming aligned" functions.
54 void addCharactersAssumingAligned(UChar a, UChar b)
56 ASSERT(!m_hasPendingCharacter);
58 m_hash = (m_hash << 16) ^ ((b << 11) ^ m_hash);
59 m_hash += m_hash >> 11;
62 void addCharacter(UChar character)
64 if (m_hasPendingCharacter) {
65 m_hasPendingCharacter = false;
66 addCharactersAssumingAligned(m_pendingCharacter, character);
70 m_pendingCharacter = character;
71 m_hasPendingCharacter = true;
74 void addCharacters(UChar a, UChar b)
76 if (m_hasPendingCharacter) {
78 m_hasPendingCharacter = false;
80 addCharactersAssumingAligned(m_pendingCharacter, a);
81 m_pendingCharacter = b;
83 m_hasPendingCharacter = true;
88 addCharactersAssumingAligned(a, b);
91 template<typename T, UChar Converter(T)> void addCharactersAssumingAligned(const T* data, unsigned length)
93 ASSERT(!m_hasPendingCharacter);
95 bool remainder = length & 1;
99 addCharactersAssumingAligned(Converter(data[0]), Converter(data[1]));
104 addCharacter(Converter(*data));
107 template<typename T> void addCharactersAssumingAligned(const T* data, unsigned length)
109 addCharactersAssumingAligned<T, defaultConverter>(data, length);
112 template<typename T, UChar Converter(T)> void addCharactersAssumingAligned(const T* data)
114 ASSERT(!m_hasPendingCharacter);
116 while (T a = *data++) {
119 addCharacter(Converter(a));
122 addCharactersAssumingAligned(Converter(a), Converter(b));
126 template<typename T> void addCharactersAssumingAligned(const T* data)
128 addCharactersAssumingAligned<T, defaultConverter>(data);
131 template<typename T, UChar Converter(T)> void addCharacters(const T* data, unsigned length)
133 if (m_hasPendingCharacter && length) {
134 m_hasPendingCharacter = false;
135 addCharactersAssumingAligned(m_pendingCharacter, Converter(*data++));
138 addCharactersAssumingAligned<T, Converter>(data, length);
141 template<typename T> void addCharacters(const T* data, unsigned length)
143 addCharacters<T, defaultConverter>(data, length);
146 template<typename T, UChar Converter(T)> void addCharacters(const T* data)
148 if (m_hasPendingCharacter && *data) {
149 m_hasPendingCharacter = false;
150 addCharactersAssumingAligned(m_pendingCharacter, Converter(*data++));
152 addCharactersAssumingAligned<T, Converter>(data);
155 template<typename T> void addCharacters(const T* data)
157 addCharacters<T, defaultConverter>(data);
160 unsigned hashWithTop8BitsMasked() const
162 unsigned result = avalancheBits();
164 // Reserving space from the high bits for flags preserves most of the hash's
165 // value, since hash lookup typically masks out the high bits anyway.
166 result &= (1U << (sizeof(result) * 8 - flagCount)) - 1;
168 // This avoids ever returning a hash code of 0, since that is used to
169 // signal "hash not computed yet". Setting the high bit maintains
170 // reasonable fidelity to a hash code of 0 because it is likely to yield
171 // exactly 0 when hash lookup masks out the high bits.
173 result = 0x80000000 >> flagCount;
178 unsigned hash() const
180 unsigned result = avalancheBits();
182 // This avoids ever returning a hash code of 0, since that is used to
183 // signal "hash not computed yet". Setting the high bit maintains
184 // reasonable fidelity to a hash code of 0 because it is likely to yield
185 // exactly 0 when hash lookup masks out the high bits.
192 template<typename T, UChar Converter(T)> static unsigned computeHashAndMaskTop8Bits(const T* data, unsigned length)
195 hasher.addCharactersAssumingAligned<T, Converter>(data, length);
196 return hasher.hashWithTop8BitsMasked();
199 template<typename T, UChar Converter(T)> static unsigned computeHashAndMaskTop8Bits(const T* data)
202 hasher.addCharactersAssumingAligned<T, Converter>(data);
203 return hasher.hashWithTop8BitsMasked();
206 template<typename T> static unsigned computeHashAndMaskTop8Bits(const T* data, unsigned length)
208 return computeHashAndMaskTop8Bits<T, defaultConverter>(data, length);
211 template<typename T> static unsigned computeHashAndMaskTop8Bits(const T* data)
213 return computeHashAndMaskTop8Bits<T, defaultConverter>(data);
216 template<typename T, UChar Converter(T)> static unsigned computeHash(const T* data, unsigned length)
219 hasher.addCharactersAssumingAligned<T, Converter>(data, length);
220 return hasher.hash();
223 template<typename T, UChar Converter(T)> static unsigned computeHash(const T* data)
226 hasher.addCharactersAssumingAligned<T, Converter>(data);
227 return hasher.hash();
230 template<typename T> static unsigned computeHash(const T* data, unsigned length)
232 return computeHash<T, defaultConverter>(data, length);
235 template<typename T> static unsigned computeHash(const T* data)
237 return computeHash<T, defaultConverter>(data);
240 static unsigned hashMemory(const void* data, unsigned length)
242 // FIXME: Why does this function use the version of the hash that drops the top 8 bits?
243 // We want that for all string hashing so we can use those bits in StringImpl and hash
244 // strings consistently, but I don't see why we'd want that for general memory hashing.
245 ASSERT(!(length % 2));
246 return computeHashAndMaskTop8Bits<UChar>(static_cast<const UChar*>(data), length / sizeof(UChar));
249 template<size_t length> static unsigned hashMemory(const void* data)
251 COMPILE_ASSERT(!(length % 2), length_must_be_a_multiple_of_two);
252 return hashMemory(data, length);
256 static UChar defaultConverter(UChar character)
261 static UChar defaultConverter(LChar character)
266 unsigned avalancheBits() const
268 unsigned result = m_hash;
271 if (m_hasPendingCharacter) {
272 result += m_pendingCharacter;
273 result ^= result << 11;
274 result += result >> 17;
277 // Force "avalanching" of final 31 bits.
278 result ^= result << 3;
279 result += result >> 5;
280 result ^= result << 2;
281 result += result >> 15;
282 result ^= result << 10;
288 bool m_hasPendingCharacter;
289 UChar m_pendingCharacter;
294 using WTF::StringHasher;
296 #endif // WTF_StringHasher_h