7 /* Stolen mostly from: lookup3.c, by Bob Jenkins, May 2006, Public Domain.
9 * http://burtleburtle.net/bob/c/lookup3.c
13 * hash - fast hash of an array for internal use
14 * @p: the array or pointer to first element
15 * @num: the number of elements to hash
16 * @base: the base number to roll into the hash (usually 0)
18 * The memory region pointed to by p is combined with the base to form
21 * This hash will have different results on different machines, so is
22 * only useful for internal hashes (ie. not hashes sent across the
23 * network or saved to disk).
25 * It may also change with future versions: it could even detect at runtime
26 * what the fastest hash to use is.
28 * See also: hash64, hash_stable.
31 * #include "hash/hash.h"
35 * // Simple demonstration: idential strings will have the same hash, but
36 * // two different strings will probably not.
37 * int main(int argc, char *argv[])
39 * uint32_t hash1, hash2;
42 * err(1, "Usage: %s <string1> <string2>", argv[0]);
44 * hash1 = hash(argv[1], strlen(argv[1]), 0);
45 * hash2 = hash(argv[2], strlen(argv[2]), 0);
46 * printf("Hash is %s\n", hash1 == hash2 ? "same" : "different");
50 #define hash(p, num, base) hash_any((p), (num)*sizeof(*(p)), (base))
53 * hash_stable - hash of an array for external use
54 * @p: the array or pointer to first element
55 * @num: the number of elements to hash
56 * @base: the base number to roll into the hash (usually 0)
58 * The array of simple integer types pointed to by p is combined with
59 * the base to form a 32-bit hash.
61 * This hash will have the same results on different machines, so can
62 * be used for external hashes (ie. hashes sent across the network or
63 * saved to disk). The results will not change in future versions of
66 * Note that it is only legal to hand an array of simple integer types
67 * to this hash (ie. char, uint16_t, int64_t, etc). In these cases,
68 * the same values will have the same hash result, even though the
69 * memory representations of integers depend on the machine
76 * #include "hash/hash.h"
80 * int main(int argc, char *argv[])
83 * err(1, "Usage: %s <string-to-hash>", argv[0]);
85 * printf("Hash stable result is %u\n",
86 * hash_stable(argv[1], strlen(argv[1]), 0));
90 #define hash_stable(p, num, base) \
91 (EXPR_BUILD_ASSERT(sizeof(*(p)) == 8 || sizeof(*(p)) == 4 \
92 || sizeof(*(p)) == 2 || sizeof(*(p)) == 1) + \
93 sizeof(*(p)) == 8 ? hash_stable_64((p), (num), (base)) \
94 : sizeof(*(p)) == 4 ? hash_stable_32((p), (num), (base)) \
95 : sizeof(*(p)) == 2 ? hash_stable_16((p), (num), (base)) \
96 : hash_stable_8((p), (num), (base)))
99 * hash_u32 - fast hash an array of 32-bit values for internal use
100 * @key: the array of uint32_t
101 * @num: the number of elements to hash
102 * @base: the base number to roll into the hash (usually 0)
104 * The array of uint32_t pointed to by @key is combined with the base
105 * to form a 32-bit hash. This is 2-3 times faster than hash() on small
106 * arrays, but the advantage vanishes over large hashes.
108 * This hash will have different results on different machines, so is
109 * only useful for internal hashes (ie. not hashes sent across the
110 * network or saved to disk).
112 uint32_t hash_u32(const uint32_t *key, size_t num, uint32_t base);
115 * hash_string - very fast hash of an ascii string
116 * @str: the nul-terminated string
118 * The string is hashed, using a hash function optimized for ASCII and
119 * similar strings. It's weaker than the other hash functions.
121 * This hash may have different results on different machines, so is
122 * only useful for internal hashes (ie. not hashes sent across the
123 * network or saved to disk). The results will be different from the
124 * other hash functions in this module, too.
126 static inline uint32_t hash_string(const char *string)
128 /* This is Karl Nelson <kenelson@ece.ucdavis.edu>'s X31 hash.
129 * It's a little faster than the (much better) lookup3 hash(): 56ns vs
130 * 84ns on my 2GHz Intel Core Duo 2 laptop for a 10 char string. */
133 for (ret = 0; *string; string++)
134 ret = (ret << 5) - ret + *string;
140 * hash64 - fast 64-bit hash of an array for internal use
141 * @p: the array or pointer to first element
142 * @num: the number of elements to hash
143 * @base: the 64-bit base number to roll into the hash (usually 0)
145 * The memory region pointed to by p is combined with the base to form
148 * This hash will have different results on different machines, so is
149 * only useful for internal hashes (ie. not hashes sent across the
150 * network or saved to disk).
152 * It may also change with future versions: it could even detect at runtime
153 * what the fastest hash to use is.
158 * #include <ccan/hash/hash.h>
162 * // Simple demonstration: idential strings will have the same hash, but
163 * // two different strings will probably not.
164 * int main(int argc, char *argv[])
166 * uint64_t hash1, hash2;
169 * err(1, "Usage: %s <string1> <string2>", argv[0]);
171 * hash1 = hash64(argv[1], strlen(argv[1]), 0);
172 * hash2 = hash64(argv[2], strlen(argv[2]), 0);
173 * printf("Hash is %s\n", hash1 == hash2 ? "same" : "different");
177 #define hash64(p, num, base) hash64_any((p), (num)*sizeof(*(p)), (base))
180 * hash64_stable - 64 bit hash of an array for external use
181 * @p: the array or pointer to first element
182 * @num: the number of elements to hash
183 * @base: the base number to roll into the hash (usually 0)
185 * The array of simple integer types pointed to by p is combined with
186 * the base to form a 64-bit hash.
188 * This hash will have the same results on different machines, so can
189 * be used for external hashes (ie. hashes sent across the network or
190 * saved to disk). The results will not change in future versions of
193 * Note that it is only legal to hand an array of simple integer types
194 * to this hash (ie. char, uint16_t, int64_t, etc). In these cases,
195 * the same values will have the same hash result, even though the
196 * memory representations of integers depend on the machine
203 * #include <ccan/hash/hash.h>
207 * int main(int argc, char *argv[])
210 * err(1, "Usage: %s <string-to-hash>", argv[0]);
212 * printf("Hash stable result is %llu\n",
213 * (long long)hash64_stable(argv[1], strlen(argv[1]), 0));
217 #define hash64_stable(p, num, base) \
218 (EXPR_BUILD_ASSERT(sizeof(*(p)) == 8 || sizeof(*(p)) == 4 \
219 || sizeof(*(p)) == 2 || sizeof(*(p)) == 1) + \
220 sizeof(*(p)) == 8 ? hash64_stable_64((p), (num), (base)) \
221 : sizeof(*(p)) == 4 ? hash64_stable_32((p), (num), (base)) \
222 : sizeof(*(p)) == 2 ? hash64_stable_16((p), (num), (base)) \
223 : hash64_stable_8((p), (num), (base)))
227 * hashl - fast 32/64-bit hash of an array for internal use
228 * @p: the array or pointer to first element
229 * @num: the number of elements to hash
230 * @base: the base number to roll into the hash (usually 0)
232 * This is either hash() or hash64(), on 32/64 bit long machines.
234 #define hashl(p, num, base) \
235 (EXPR_BUILD_ASSERT(sizeof(long) == sizeof(uint32_t) \
236 || sizeof(long) == sizeof(uint64_t)) + \
237 (sizeof(long) == sizeof(uint64_t) \
238 ? hash64((p), (num), (base)) : hash((p), (num), (base))))
240 /* Our underlying operations. */
241 uint32_t hash_any(const void *key, size_t length, uint32_t base);
242 uint32_t hash_stable_64(const void *key, size_t n, uint32_t base);
243 uint32_t hash_stable_32(const void *key, size_t n, uint32_t base);
244 uint32_t hash_stable_16(const void *key, size_t n, uint32_t base);
245 uint32_t hash_stable_8(const void *key, size_t n, uint32_t base);
246 uint64_t hash64_any(const void *key, size_t length, uint64_t base);
247 uint64_t hash64_stable_64(const void *key, size_t n, uint64_t base);
248 uint64_t hash64_stable_32(const void *key, size_t n, uint64_t base);
249 uint64_t hash64_stable_16(const void *key, size_t n, uint64_t base);
250 uint64_t hash64_stable_8(const void *key, size_t n, uint64_t base);
253 * hash_pointer - hash a pointer for internal use
254 * @p: the pointer value to hash
255 * @base: the base number to roll into the hash (usually 0)
257 * The pointer p (not what p points to!) is combined with the base to form
260 * This hash will have different results on different machines, so is
261 * only useful for internal hashes (ie. not hashes sent across the
262 * network or saved to disk).
265 * #include "hash/hash.h"
267 * // Code to keep track of memory regions.
269 * struct region *chain;
273 * // We keep a simple hash table.
274 * static struct region *region_hash[128];
276 * static void add_region(struct region *r)
278 * unsigned int h = hash_pointer(r->start);
280 * r->chain = region_hash[h];
281 * region_hash[h] = r->chain;
284 * static void find_region(const void *start)
288 * for (r = region_hash[hash_pointer(start)]; r; r = r->chain)
289 * if (r->start == start)
294 static inline uint32_t hash_pointer(const void *p, uint32_t base)
296 if (sizeof(p) % sizeof(uint32_t) == 0) {
297 /* This convoluted union is the right way of aliasing. */
299 uint32_t u32[sizeof(p) / sizeof(uint32_t)];
303 return hash_u32(u.u32, sizeof(p) / sizeof(uint32_t), base);
305 return hash(&p, 1, base);