tanjent@gmail.com | 510b852 | 2011-04-12 15:36:18 +0000 | [diff] [blame] | 1 | // Copyright (c) 2011 Google, Inc.
|
| 2 | //
|
| 3 | // Permission is hereby granted, free of charge, to any person obtaining a copy
|
| 4 | // of this software and associated documentation files (the "Software"), to deal
|
| 5 | // in the Software without restriction, including without limitation the rights
|
| 6 | // to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
| 7 | // copies of the Software, and to permit persons to whom the Software is
|
| 8 | // furnished to do so, subject to the following conditions:
|
| 9 | //
|
| 10 | // The above copyright notice and this permission notice shall be included in
|
| 11 | // all copies or substantial portions of the Software.
|
| 12 | //
|
| 13 | // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
| 14 | // IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
| 15 | // FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
| 16 | // AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
| 17 | // LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
| 18 | // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
| 19 | // THE SOFTWARE.
|
| 20 | //
|
| 21 | // CityHash Version 1, by Geoff Pike and Jyrki Alakuijala
|
| 22 | //
|
| 23 | // This file provides CityHash64() and related functions.
|
| 24 | //
|
| 25 | // It's probably possible to create even faster hash functions by
|
| 26 | // writing a program that systematically explores some of the space of
|
| 27 | // possible hash functions, by using SIMD instructions, or by
|
| 28 | // compromising on hash quality.
|
| 29 |
|
aappleby@google.com | a27c281 | 2011-04-13 23:23:14 +0000 | [diff] [blame] | 30 | #include "City.h"
|
tanjent@gmail.com | 510b852 | 2011-04-12 15:36:18 +0000 | [diff] [blame] | 31 |
|
| 32 | #include <algorithm>
|
| 33 |
|
| 34 | using namespace std;
|
| 35 |
|
| 36 | #define UNALIGNED_LOAD64(p) (*(const uint64*)(p))
|
| 37 | #define UNALIGNED_LOAD32(p) (*(const uint32*)(p))
|
| 38 |
|
| 39 | #if !defined(LIKELY)
|
| 40 | #if defined(__GNUC__)
|
| 41 | #define LIKELY(x) (__builtin_expect(!!(x), 1))
|
| 42 | #else
|
| 43 | #define LIKELY(x) (x)
|
| 44 | #endif
|
| 45 | #endif
|
| 46 |
|
| 47 | // Some primes between 2^63 and 2^64 for various uses.
|
| 48 | static const uint64 k0 = 0xc3a5c85c97cb3127ULL;
|
| 49 | static const uint64 k1 = 0xb492b66fbe98f273ULL;
|
| 50 | static const uint64 k2 = 0x9ae16a3b2f90404fULL;
|
| 51 | static const uint64 k3 = 0xc949d7c7509e6557ULL;
|
| 52 |
|
| 53 | // Bitwise right rotate. Normally this will compile to a single
|
| 54 | // instruction, especially if the shift is a manifest constant.
|
| 55 | static uint64 Rotate(uint64 val, int shift) {
|
| 56 | // Avoid shifting by 64: doing so yields an undefined result.
|
| 57 | return shift == 0 ? val : ((val >> shift) | (val << (64 - shift)));
|
| 58 | }
|
| 59 |
|
| 60 | // Equivalent to Rotate(), but requires the second arg to be non-zero.
|
| 61 | // On x86-64, and probably others, it's possible for this to compile
|
| 62 | // to a single instruction if both args are already in registers.
|
| 63 | static uint64 RotateByAtLeast1(uint64 val, int shift) {
|
| 64 | return (val >> shift) | (val << (64 - shift));
|
| 65 | }
|
| 66 |
|
| 67 | static uint64 ShiftMix(uint64 val) {
|
| 68 | return val ^ (val >> 47);
|
| 69 | }
|
| 70 |
|
| 71 | static uint64 HashLen16(uint64 u, uint64 v) {
|
| 72 | return Hash128to64(uint128(u, v));
|
| 73 | }
|
| 74 |
|
| 75 | static uint64 HashLen0to16(const char *s, size_t len) {
|
| 76 | if (len > 8) {
|
| 77 | uint64 a = UNALIGNED_LOAD64(s);
|
| 78 | uint64 b = UNALIGNED_LOAD64(s + len - 8);
|
| 79 | return HashLen16(a, RotateByAtLeast1(b + len, len)) ^ b;
|
| 80 | }
|
| 81 | if (len >= 4) {
|
| 82 | uint64 a = UNALIGNED_LOAD32(s);
|
| 83 | return HashLen16(len + (a << 3), UNALIGNED_LOAD32(s + len - 4));
|
| 84 | }
|
| 85 | if (len > 0) {
|
| 86 | uint8 a = s[0];
|
| 87 | uint8 b = s[len >> 1];
|
| 88 | uint8 c = s[len - 1];
|
| 89 | uint32 y = static_cast<uint32>(a) + (static_cast<uint32>(b) << 8);
|
| 90 | uint32 z = len + (static_cast<uint32>(c) << 2);
|
| 91 | return ShiftMix(y * k2 ^ z * k3) * k2;
|
| 92 | }
|
| 93 | return k2;
|
| 94 | }
|
| 95 |
|
| 96 | // This probably works well for 16-byte strings as well, but it may be overkill
|
| 97 | // in that case.
|
| 98 | static uint64 HashLen17to32(const char *s, size_t len) {
|
| 99 | uint64 a = UNALIGNED_LOAD64(s) * k1;
|
| 100 | uint64 b = UNALIGNED_LOAD64(s + 8);
|
| 101 | uint64 c = UNALIGNED_LOAD64(s + len - 8) * k2;
|
| 102 | uint64 d = UNALIGNED_LOAD64(s + len - 16) * k0;
|
| 103 | return HashLen16(Rotate(a - b, 43) + Rotate(c, 30) + d,
|
| 104 | a + Rotate(b ^ k3, 20) - c + len);
|
| 105 | }
|
| 106 |
|
| 107 | // Return a 16-byte hash for 48 bytes. Quick and dirty.
|
| 108 | // Callers do best to use "random-looking" values for a and b.
|
| 109 | static pair<uint64, uint64> WeakHashLen32WithSeeds(
|
| 110 | uint64 w, uint64 x, uint64 y, uint64 z, uint64 a, uint64 b) {
|
| 111 | a += w;
|
| 112 | b = Rotate(b + a + z, 21);
|
| 113 | uint64 c = a;
|
| 114 | a += x;
|
| 115 | a += y;
|
| 116 | b += Rotate(a, 44);
|
| 117 | return make_pair(a + z, b + c);
|
| 118 | }
|
| 119 |
|
| 120 | // Return a 16-byte hash for s[0] ... s[31], a, and b. Quick and dirty.
|
| 121 | static pair<uint64, uint64> WeakHashLen32WithSeeds(
|
| 122 | const char* s, uint64 a, uint64 b) {
|
| 123 | return WeakHashLen32WithSeeds(UNALIGNED_LOAD64(s),
|
| 124 | UNALIGNED_LOAD64(s + 8),
|
| 125 | UNALIGNED_LOAD64(s + 16),
|
| 126 | UNALIGNED_LOAD64(s + 24),
|
| 127 | a,
|
| 128 | b);
|
| 129 | }
|
| 130 |
|
| 131 | // Return an 8-byte hash for 33 to 64 bytes.
|
| 132 | static uint64 HashLen33to64(const char *s, size_t len) {
|
| 133 | uint64 z = UNALIGNED_LOAD64(s + 24);
|
| 134 | uint64 a = UNALIGNED_LOAD64(s) + (len + UNALIGNED_LOAD64(s + len - 16)) * k0;
|
| 135 | uint64 b = Rotate(a + z, 52);
|
| 136 | uint64 c = Rotate(a, 37);
|
| 137 | a += UNALIGNED_LOAD64(s + 8);
|
| 138 | c += Rotate(a, 7);
|
| 139 | a += UNALIGNED_LOAD64(s + 16);
|
| 140 | uint64 vf = a + z;
|
| 141 | uint64 vs = b + Rotate(a, 31) + c;
|
| 142 | a = UNALIGNED_LOAD64(s + 16) + UNALIGNED_LOAD64(s + len - 32);
|
| 143 | z = UNALIGNED_LOAD64(s + len - 8);
|
| 144 | b = Rotate(a + z, 52);
|
| 145 | c = Rotate(a, 37);
|
| 146 | a += UNALIGNED_LOAD64(s + len - 24);
|
| 147 | c += Rotate(a, 7);
|
| 148 | a += UNALIGNED_LOAD64(s + len - 16);
|
| 149 | uint64 wf = a + z;
|
| 150 | uint64 ws = b + Rotate(a, 31) + c;
|
| 151 | uint64 r = ShiftMix((vf + ws) * k2 + (wf + vs) * k0);
|
| 152 | return ShiftMix(r * k0 + vs) * k2;
|
| 153 | }
|
| 154 |
|
| 155 | uint64 CityHash64(const char *s, size_t len) {
|
| 156 | if (len <= 32) {
|
| 157 | if (len <= 16) {
|
| 158 | return HashLen0to16(s, len);
|
| 159 | } else {
|
| 160 | return HashLen17to32(s, len);
|
| 161 | }
|
| 162 | } else if (len <= 64) {
|
| 163 | return HashLen33to64(s, len);
|
| 164 | }
|
| 165 |
|
| 166 | // For strings over 64 bytes we hash the end first, and then as we
|
| 167 | // loop we keep 56 bytes of state: v, w, x, y, and z.
|
| 168 | uint64 x = UNALIGNED_LOAD64(s);
|
| 169 | uint64 y = UNALIGNED_LOAD64(s + len - 16) ^ k1;
|
| 170 | uint64 z = UNALIGNED_LOAD64(s + len - 56) ^ k0;
|
| 171 | pair<uint64, uint64> v = WeakHashLen32WithSeeds(s + len - 64, len, y);
|
| 172 | pair<uint64, uint64> w = WeakHashLen32WithSeeds(s + len - 32, len * k1, k0);
|
| 173 | z += ShiftMix(v.second) * k1;
|
| 174 | x = Rotate(z + x, 39) * k1;
|
| 175 | y = Rotate(y, 33) * k1;
|
| 176 |
|
| 177 | // Decrease len to the nearest multiple of 64, and operate on 64-byte chunks.
|
| 178 | len = (len - 1) & ~static_cast<size_t>(63);
|
| 179 | do {
|
| 180 | x = Rotate(x + y + v.first + UNALIGNED_LOAD64(s + 16), 37) * k1;
|
| 181 | y = Rotate(y + v.second + UNALIGNED_LOAD64(s + 48), 42) * k1;
|
| 182 | x ^= w.second;
|
| 183 | y ^= v.first;
|
| 184 | z = Rotate(z ^ w.first, 33);
|
| 185 | v = WeakHashLen32WithSeeds(s, v.second * k1, x + w.first);
|
| 186 | w = WeakHashLen32WithSeeds(s + 32, z + w.second, y);
|
| 187 | std::swap(z, x);
|
| 188 | s += 64;
|
| 189 | len -= 64;
|
| 190 | } while (len != 0);
|
| 191 | return HashLen16(HashLen16(v.first, w.first) + ShiftMix(y) * k1 + z,
|
| 192 | HashLen16(v.second, w.second) + x);
|
| 193 | }
|
| 194 |
|
| 195 | uint64 CityHash64WithSeed(const char *s, size_t len, uint64 seed) {
|
| 196 | return CityHash64WithSeeds(s, len, k2, seed);
|
| 197 | }
|
| 198 |
|
| 199 | uint64 CityHash64WithSeeds(const char *s, size_t len,
|
| 200 | uint64 seed0, uint64 seed1) {
|
| 201 | return HashLen16(CityHash64(s, len) - seed0, seed1);
|
| 202 | }
|
| 203 |
|
| 204 | // A subroutine for CityHash128(). Returns a decent 128-bit hash for strings
|
| 205 | // of any length representable in ssize_t. Based on City and Murmur.
|
| 206 | static uint128 CityMurmur(const char *s, size_t len, uint128 seed) {
|
| 207 | uint64 a = Uint128Low64(seed);
|
| 208 | uint64 b = Uint128High64(seed);
|
| 209 | uint64 c = 0;
|
| 210 | uint64 d = 0;
|
| 211 | ssize_t l = len - 16;
|
| 212 | if (l <= 0) { // len <= 16
|
| 213 | c = b * k1 + HashLen0to16(s, len);
|
| 214 | d = Rotate(a + (len >= 8 ? UNALIGNED_LOAD64(s) : c), 32);
|
| 215 | } else { // len > 16
|
| 216 | c = HashLen16(UNALIGNED_LOAD64(s + len - 8) + k1, a);
|
| 217 | d = HashLen16(b + len, c + UNALIGNED_LOAD64(s + len - 16));
|
| 218 | a += d;
|
| 219 | do {
|
| 220 | a ^= ShiftMix(UNALIGNED_LOAD64(s) * k1) * k1;
|
| 221 | a *= k1;
|
| 222 | b ^= a;
|
| 223 | c ^= ShiftMix(UNALIGNED_LOAD64(s + 8) * k1) * k1;
|
| 224 | c *= k1;
|
| 225 | d ^= c;
|
| 226 | s += 16;
|
| 227 | l -= 16;
|
| 228 | } while (l > 0);
|
| 229 | }
|
| 230 | a = HashLen16(a, c);
|
| 231 | b = HashLen16(d, b);
|
| 232 | return uint128(a ^ b, HashLen16(b, a));
|
| 233 | }
|
| 234 |
|
| 235 | uint128 CityHash128WithSeed(const char *s, size_t len, uint128 seed) {
|
| 236 | if (len < 128) {
|
| 237 | return CityMurmur(s, len, seed);
|
| 238 | }
|
| 239 |
|
| 240 | // We expect len >= 128 to be the common case. Keep 56 bytes of state:
|
| 241 | // v, w, x, y, and z.
|
| 242 | pair<uint64, uint64> v, w;
|
| 243 | uint64 x = Uint128Low64(seed);
|
| 244 | uint64 y = Uint128High64(seed);
|
| 245 | uint64 z = len * k1;
|
| 246 | v.first = Rotate(y ^ k1, 49) * k1 + UNALIGNED_LOAD64(s);
|
| 247 | v.second = Rotate(v.first, 42) * k1 + UNALIGNED_LOAD64(s + 8);
|
| 248 | w.first = Rotate(y + z, 35) * k1 + x;
|
| 249 | w.second = Rotate(x + UNALIGNED_LOAD64(s + 88), 53) * k1;
|
| 250 |
|
| 251 | // This is the same inner loop as CityHash64(), manually unrolled.
|
| 252 | do {
|
| 253 | x = Rotate(x + y + v.first + UNALIGNED_LOAD64(s + 16), 37) * k1;
|
| 254 | y = Rotate(y + v.second + UNALIGNED_LOAD64(s + 48), 42) * k1;
|
| 255 | x ^= w.second;
|
| 256 | y ^= v.first;
|
| 257 | z = Rotate(z ^ w.first, 33);
|
| 258 | v = WeakHashLen32WithSeeds(s, v.second * k1, x + w.first);
|
| 259 | w = WeakHashLen32WithSeeds(s + 32, z + w.second, y);
|
| 260 | std::swap(z, x);
|
| 261 | s += 64;
|
| 262 | x = Rotate(x + y + v.first + UNALIGNED_LOAD64(s + 16), 37) * k1;
|
| 263 | y = Rotate(y + v.second + UNALIGNED_LOAD64(s + 48), 42) * k1;
|
| 264 | x ^= w.second;
|
| 265 | y ^= v.first;
|
| 266 | z = Rotate(z ^ w.first, 33);
|
| 267 | v = WeakHashLen32WithSeeds(s, v.second * k1, x + w.first);
|
| 268 | w = WeakHashLen32WithSeeds(s + 32, z + w.second, y);
|
| 269 | std::swap(z, x);
|
| 270 | s += 64;
|
| 271 | len -= 128;
|
| 272 | } while (LIKELY(len >= 128));
|
| 273 | y += Rotate(w.first, 37) * k0 + z;
|
| 274 | x += Rotate(v.first + z, 49) * k0;
|
| 275 | // If 0 < len < 128, hash up to 4 chunks of 32 bytes each from the end of s.
|
| 276 | for (size_t tail_done = 0; tail_done < len; ) {
|
| 277 | tail_done += 32;
|
| 278 | y = Rotate(y - x, 42) * k0 + v.second;
|
| 279 | w.first += UNALIGNED_LOAD64(s + len - tail_done + 16);
|
| 280 | x = Rotate(x, 49) * k0 + w.first;
|
| 281 | w.first += v.first;
|
| 282 | v = WeakHashLen32WithSeeds(s + len - tail_done, v.first, v.second);
|
| 283 | }
|
| 284 | // At this point our 48 bytes of state should contain more than
|
| 285 | // enough information for a strong 128-bit hash. We use two
|
| 286 | // different 48-byte-to-8-byte hashes to get a 16-byte final result.
|
| 287 | x = HashLen16(x, v.first);
|
| 288 | y = HashLen16(y, w.first);
|
| 289 | return uint128(HashLen16(x + v.second, w.second) + y,
|
| 290 | HashLen16(x + w.second, y + v.second));
|
| 291 | }
|
| 292 |
|
| 293 | uint128 CityHash128(const char *s, size_t len) {
|
| 294 | if (len >= 16) {
|
| 295 | return CityHash128WithSeed(s + 16,
|
| 296 | len - 16,
|
| 297 | uint128(UNALIGNED_LOAD64(s) ^ k3,
|
| 298 | UNALIGNED_LOAD64(s + 8)));
|
| 299 | } else if (len >= 8) {
|
| 300 | return CityHash128WithSeed(NULL,
|
| 301 | 0,
|
| 302 | uint128(UNALIGNED_LOAD64(s) ^ (len * k0),
|
| 303 | UNALIGNED_LOAD64(s + len - 8) ^ k1));
|
| 304 | } else {
|
| 305 | return CityHash128WithSeed(s, len, uint128(k0, k1));
|
| 306 | }
|
| 307 | }
|
| 308 |
|
| 309 | void CityHash64_test ( const void * key, int len, uint32_t seed, void * out )
|
| 310 | {
|
| 311 | *(uint64*)out = CityHash64WithSeed((const char *)key,len,seed);
|
| 312 | }
|
| 313 |
|
| 314 | void CityHash128_test ( const void * key, int len, uint32_t seed, void * out )
|
| 315 | {
|
| 316 | uint128 s(0,0);
|
| 317 |
|
| 318 | s.first = seed;
|
| 319 |
|
| 320 | *(uint128*)out = CityHash128WithSeed((const char*)key,len,s);
|
aappleby@google.com | a27c281 | 2011-04-13 23:23:14 +0000 | [diff] [blame] | 321 | }
|