2 Copyright (C) 2001-2006, William Joseph.
5 This file is part of GtkRadiant.
7 GtkRadiant is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 GtkRadiant is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with GtkRadiant; if not, write to the Free Software
19 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
22 #if !defined(INCLUDED_CONTAINER_HASHFUNC_H)
23 #define INCLUDED_CONTAINER_HASHFUNC_H
26 #include "string/string.h"
27 #include "container/array.h"
28 typedef unsigned long int ub4; /* unsigned 4-byte quantities */
29 typedef unsigned char ub1;
31 inline ub1 ub1_as_ub1_nocase(ub1 byte)
33 return std::tolower(byte);
36 inline ub4 ub1x4_as_ub4_nocase(const ub1 bytes[4])
39 reinterpret_cast<ub1*>(&result)[0] = ub1_as_ub1_nocase(bytes[0]);
40 reinterpret_cast<ub1*>(&result)[1] = ub1_as_ub1_nocase(bytes[1]);
41 reinterpret_cast<ub1*>(&result)[2] = ub1_as_ub1_nocase(bytes[2]);
42 reinterpret_cast<ub1*>(&result)[3] = ub1_as_ub1_nocase(bytes[3]);
46 class ub1_default_traits
49 static ub1 as_ub1(ub1 byte)
55 class ub1_nocase_traits
58 static ub1 as_ub1(ub1 byte)
60 return ub1_as_ub1_nocase(byte);
64 class ub1x4_default_traits
67 static ub4 as_ub4(const ub1 bytes[4])
69 return *reinterpret_cast<const ub4*>(bytes);
73 class ub1x4_nocase_traits
76 static ub4 as_ub4(const ub1 bytes[4])
78 return ub1x4_as_ub4_nocase(bytes);
82 class ub4_default_traits
85 static ub4 as_ub4(ub4 i)
91 class ub4_nocase_traits
94 static ub4 as_ub4(ub4 i)
96 return ub1x4_as_ub4_nocase(reinterpret_cast<const ub1*>(&i));
101 // By Bob Jenkins, 1996. bob_jenkins@burtleburtle.net. You may use this
102 // code any way you wish, private, educational, or commercial. It's free.
104 #define hashsize(n) ((ub4)1<<(n))
105 #define hashmask(n) (hashsize(n)-1)
108 --------------------------------------------------------------------
109 mix -- mix 3 32-bit values reversibly.
110 For every delta with one or two bit set, and the deltas of all three
111 high bits or all three low bits, whether the original value of a,b,c
112 is almost all zero or is uniformly distributed,
113 * If mix() is run forward or backward, at least 32 bits in a,b,c
114 have at least 1/4 probability of changing.
115 * If mix() is run forward, every bit of c will change between 1/3 and
116 2/3 of the time. (Well, 22/100 and 78/100 for some 2-bit deltas.)
117 mix() was built out of 36 single-cycle latency instructions in a
118 structure that could supported 2x parallelism, like so:
126 Unfortunately, superscalar Pentiums and Sparcs can't take advantage
127 of that parallelism. They've also turned some of those single-cycle
128 latency instructions into multi-cycle latency instructions. Still,
129 this is the fastest good hash I could find. There were about 2^^68
130 to choose from. I only looked at a billion or so.
131 --------------------------------------------------------------------
135 a -= b; a -= c; a ^= (c>>13); \
136 b -= c; b -= a; b ^= (a<<8); \
137 c -= a; c -= b; c ^= (b>>13); \
138 a -= b; a -= c; a ^= (c>>12); \
139 b -= c; b -= a; b ^= (a<<16); \
140 c -= a; c -= b; c ^= (b>>5); \
141 a -= b; a -= c; a ^= (c>>3); \
142 b -= c; b -= a; b ^= (a<<10); \
143 c -= a; c -= b; c ^= (b>>15); \
146 /* same, but slower, works on systems that might have 8 byte ub4's */
147 #define mix2(a,b,c) \
149 a -= b; a -= c; a ^= (c>>13); \
150 b -= c; b -= a; b ^= (a<< 8); \
151 c -= a; c -= b; c ^= ((b&0xffffffff)>>13); \
152 a -= b; a -= c; a ^= ((c&0xffffffff)>>12); \
153 b -= c; b -= a; b = (b ^ (a<<16)) & 0xffffffff; \
154 c -= a; c -= b; c = (c ^ (b>> 5)) & 0xffffffff; \
155 a -= b; a -= c; a = (a ^ (c>> 3)) & 0xffffffff; \
156 b -= c; b -= a; b = (b ^ (a<<10)) & 0xffffffff; \
157 c -= a; c -= b; c = (c ^ (b>>15)) & 0xffffffff; \
161 --------------------------------------------------------------------
162 hash() -- hash a variable-length key into a 32-bit value
163 k : the key (the unaligned variable-length array of bytes)
164 len : the length of the key, counting by bytes
165 level : can be any 4-byte value
166 Returns a 32-bit value. Every bit of the key affects every bit of
167 the return value. Every 1-bit and 2-bit delta achieves avalanche.
168 About 36+6len instructions.
170 The best hash table sizes are powers of 2. There is no need to do
171 mod a prime (mod is sooo slow!). If you need less than 32 bits,
172 use a bitmask. For example, if you need only 10 bits, do
173 h = (h & hashmask(10));
174 In which case, the hash table should have hashsize(10) elements.
176 If you are hashing n strings (ub1 **)k, do it like this:
177 for (i=0, h=0; i<n; ++i) h = hash( k[i], len[i], h);
179 See http://burlteburtle.net/bob/hash/evahash.html
180 Use for hash table lookup, or anything where one collision in 2^32 is
181 acceptable. Do NOT use for cryptographic purposes.
182 --------------------------------------------------------------------
185 template<typename UB1Traits, typename UB4x1Traits>
187 const ub1 *k, /* the key */
188 ub4 length, /* the length of the key */
189 ub4 initval, /* the previous hash, or an arbitrary value */
190 const UB1Traits& ub1traits,
191 const UB4x1Traits& ub4x1traits
194 register ub4 a,b,c,len;
196 /* Set up the internal state */
198 a = b = 0x9e3779b9; /* the golden ratio; an arbitrary value */
199 c = initval; /* the previous hash value */
201 /*---------------------------------------- handle most of the key */
204 a += (k[0] +((ub4)UB1Traits::as_ub1(k[1])<<8) +((ub4)UB1Traits::as_ub1(k[2])<<16) +((ub4)UB1Traits::as_ub1(k[3])<<24));
205 b += (k[4] +((ub4)UB1Traits::as_ub1(k[5])<<8) +((ub4)UB1Traits::as_ub1(k[6])<<16) +((ub4)UB1Traits::as_ub1(k[7])<<24));
206 c += (k[8] +((ub4)UB1Traits::as_ub1(k[9])<<8) +((ub4)UB1Traits::as_ub1(k[10])<<16)+((ub4)UB1Traits::as_ub1(k[11])<<24));
211 /*------------------------------------- handle the last 11 bytes */
213 switch(len) /* all the case statements fall through */
215 case 11: c += ((ub4)UB1Traits::as_ub1(k[10]) << 24);
216 case 10: c += ((ub4)UB1Traits::as_ub1(k[9]) << 16);
217 case 9 : c += ((ub4)UB1Traits::as_ub1(k[8]) << 8);
218 /* the first byte of c is reserved for the length */
219 case 8 : b += ((ub4)UB1Traits::as_ub1(k[7]) << 24);
220 case 7 : b += ((ub4)UB1Traits::as_ub1(k[6]) << 16);
221 case 6 : b += ((ub4)UB1Traits::as_ub1(k[5]) << 8);
222 case 5 : b += UB1Traits::as_ub1(k[4]);
223 case 4 : a += ((ub4)UB1Traits::as_ub1(k[3]) << 24);
224 case 3 : a += ((ub4)UB1Traits::as_ub1(k[2]) << 16);
225 case 2 : a += ((ub4)UB1Traits::as_ub1(k[1]) << 8);
226 case 1 : a += UB1Traits::as_ub1(k[0]);
227 /* case 0: nothing left to add */
230 /*-------------------------------------------- report the result */
235 --------------------------------------------------------------------
236 This works on all machines. hash2() is identical to hash() on
237 little-endian machines, except that the length has to be measured
238 in ub4s instead of bytes. It is much faster than hash(). It
240 -- that the key be an array of ub4's, and
241 -- that all your machines have the same endianness, and
242 -- that the length be the number of ub4's in the key
243 --------------------------------------------------------------------
245 template<typename UB4Traits>
247 const ub4 *k, /* the key */
248 ub4 length, /* the length of the key, in ub4s */
249 ub4 initval, /* the previous hash, or an arbitrary value */
250 const UB4Traits& ub4traits
253 register ub4 a,b,c,len;
255 /* Set up the internal state */
257 a = b = 0x9e3779b9; /* the golden ratio; an arbitrary value */
258 c = initval; /* the previous hash value */
260 /*---------------------------------------- handle most of the key */
263 a += UB4Traits::as_ub4(k[0]);
264 b += UB4Traits::as_ub4(k[1]);
265 c += UB4Traits::as_ub4(k[2]);
270 /*-------------------------------------- handle the last 2 ub4's */
272 switch(len) /* all the case statements fall through */
274 /* c is reserved for the length */
275 case 2 : b += UB4Traits::as_ub4(k[1]);
276 case 1 : a += UB4Traits::as_ub4(k[0]);
277 /* case 0: nothing left to add */
280 /*-------------------------------------------- report the result */
286 inline hash_t hash_ub1(const ub1* key, std::size_t len, hash_t previous = 0)
288 return hash(key, ub4(len), previous, ub1_default_traits(), ub1x4_default_traits());
291 inline hash_t hash_ub1_nocase(const ub1* key, std::size_t len, hash_t previous = 0)
293 return hash(key, ub4(len), previous, ub1_nocase_traits(), ub1x4_nocase_traits());
296 template<typename UB4Traits>
297 inline hash_t hash_ub4(const ub4* key, std::size_t len, const UB4Traits& traits, hash_t previous = 0)
299 return hash2(key,ub4(len), previous, traits);
302 inline ub4 hash_combine(ub4 left, ub4 right)
304 return hash_ub1(reinterpret_cast<const ub1*>(&left), 4, right);
307 template<typename POD>
308 inline hash_t pod_hash(const POD& pod)
310 return hash_ub1(reinterpret_cast<const ub1*>(&pod), sizeof(POD));
313 inline hash_t string_hash(const char* string, hash_t previous = 0)
315 return hash_ub1(reinterpret_cast<const ub1*>(string), string_length(string), previous);
318 inline hash_t string_hash_nocase(const char* string, hash_t previous = 0)
320 return hash_ub1_nocase(reinterpret_cast<const ub1*>(string), string_length(string), previous);
325 typedef hash_t hash_type;
326 hash_type operator()(const CopiedString& string) const
328 return string_hash(string.c_str());
332 struct HashStringNoCase
334 typedef hash_t hash_type;
335 hash_type operator()(const CopiedString& string) const
337 return string_hash_nocase(string.c_str());
341 /// \brief Length of a string in ub4.
342 /// "wibble" (6) gives 2,
343 /// "and" (3) gives 1,
344 /// "bleh" (4) gives 2
345 inline std::size_t string_length_ub4(const char* string)
347 return ((string_length(string)>>2)+1)<<2;
350 /// \brief Hashable key type that stores a string as an array of ub4 - making hashing faster.
351 /// Also caches the 32-bit result of the hash to speed up comparison of keys.
352 template<typename UB4Traits = ub4_default_traits>
358 void copy(const HashKey& other)
360 std::copy(other.m_key.begin(), other.m_key.end(), m_key.begin());
361 m_hash = other.m_hash;
363 void copy(const char* string)
365 strncpy(reinterpret_cast<char*>(m_key.data()), string, m_key.size());
366 for(Array<ub4>::iterator i = m_key.begin(); i != m_key.end(); ++i)
368 *i = UB4Traits::as_ub4(*i);
370 m_hash = hash_ub4(m_key.data(), m_key.size(), ub4_default_traits());
372 bool equal(const HashKey& other) const
374 return m_hash == other.m_hash && m_key.size() == other.m_key.size()
375 && std::equal(m_key.begin(), m_key.end(), other.m_key.begin());
379 HashKey(const HashKey& other) : m_key(other.m_key.size())
383 HashKey(const char* string) : m_key(string_length_ub4(string))
387 HashKey& operator=(const char* string)
389 m_key.resize(string_length_ub4(string));
393 bool operator==(const HashKey& other) const
397 bool operator!=(const HashKey& other) const
399 return !equal(other);
406 const char* c_str() const
408 return reinterpret_cast<const char*>(m_key.data());
413 /// \brief Hash function to use with HashKey.
416 typedef hash_t hash_type;
417 hash_type operator()(const HashKey<ub4_default_traits>& key) const