tanjent@gmail.com | f3b7897 | 2012-03-01 03:38:55 +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, 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 | |
| 30 | #include "City.h" |
| 31 | |
| 32 | #include <algorithm> |
| 33 | #include <string.h> // for memcpy and memset |
| 34 | |
| 35 | using namespace std; |
| 36 | |
| 37 | static uint64 UNALIGNED_LOAD64(const char *p) { |
| 38 | uint64 result; |
| 39 | memcpy(&result, p, sizeof(result)); |
| 40 | return result; |
| 41 | } |
| 42 | |
| 43 | static uint32 UNALIGNED_LOAD32(const char *p) { |
| 44 | uint32 result; |
| 45 | memcpy(&result, p, sizeof(result)); |
| 46 | return result; |
| 47 | } |
| 48 | |
| 49 | #ifndef __BIG_ENDIAN__ |
| 50 | |
| 51 | #define uint32_in_expected_order(x) (x) |
| 52 | #define uint64_in_expected_order(x) (x) |
| 53 | |
| 54 | #else |
| 55 | |
| 56 | #ifdef _MSC_VER |
| 57 | #include <stdlib.h> |
| 58 | #define bswap_32(x) _byteswap_ulong(x) |
| 59 | #define bswap_64(x) _byteswap_uint64(x) |
| 60 | |
| 61 | #elif defined(__APPLE__) |
| 62 | // Mac OS X / Darwin features |
| 63 | #include <libkern/OSByteOrder.h> |
| 64 | #define bswap_32(x) OSSwapInt32(x) |
| 65 | #define bswap_64(x) OSSwapInt64(x) |
| 66 | |
| 67 | #else |
| 68 | #include <byteswap.h> |
| 69 | #endif |
| 70 | |
| 71 | #define uint32_in_expected_order(x) (bswap_32(x)) |
| 72 | #define uint64_in_expected_order(x) (bswap_64(x)) |
| 73 | |
| 74 | #endif // __BIG_ENDIAN__ |
| 75 | |
| 76 | #if !defined(LIKELY) |
| 77 | #if defined(__GNUC__) || defined(__INTEL_COMPILER) |
| 78 | #define LIKELY(x) (__builtin_expect(!!(x), 1)) |
| 79 | #else |
| 80 | #define LIKELY(x) (x) |
| 81 | #endif |
| 82 | #endif |
| 83 | |
| 84 | static uint64 Fetch64(const char *p) { |
| 85 | return uint64_in_expected_order(UNALIGNED_LOAD64(p)); |
| 86 | } |
| 87 | |
| 88 | static uint32 Fetch32(const char *p) { |
| 89 | return uint32_in_expected_order(UNALIGNED_LOAD32(p)); |
| 90 | } |
| 91 | |
| 92 | // Some primes between 2^63 and 2^64 for various uses. |
| 93 | static const uint64 k0 = 0xc3a5c85c97cb3127ULL; |
| 94 | static const uint64 k1 = 0xb492b66fbe98f273ULL; |
| 95 | static const uint64 k2 = 0x9ae16a3b2f90404fULL; |
| 96 | static const uint64 k3 = 0xc949d7c7509e6557ULL; |
| 97 | |
| 98 | // Bitwise right rotate. Normally this will compile to a single |
| 99 | // instruction, especially if the shift is a manifest constant. |
| 100 | static uint64 Rotate(uint64 val, int shift) { |
| 101 | // Avoid shifting by 64: doing so yields an undefined result. |
| 102 | return shift == 0 ? val : ((val >> shift) | (val << (64 - shift))); |
| 103 | } |
| 104 | |
| 105 | // Equivalent to Rotate(), but requires the second arg to be non-zero. |
| 106 | // On x86-64, and probably others, it's possible for this to compile |
| 107 | // to a single instruction if both args are already in registers. |
| 108 | static uint64 RotateByAtLeast1(uint64 val, int shift) { |
| 109 | return (val >> shift) | (val << (64 - shift)); |
| 110 | } |
| 111 | |
| 112 | static uint64 ShiftMix(uint64 val) { |
| 113 | return val ^ (val >> 47); |
| 114 | } |
| 115 | |
| 116 | static uint64 HashLen16(uint64 u, uint64 v) { |
| 117 | return Hash128to64(uint128(u, v)); |
| 118 | } |
| 119 | |
| 120 | static uint64 HashLen0to16(const char *s, size_t len) { |
| 121 | if (len > 8) { |
| 122 | uint64 a = Fetch64(s); |
| 123 | uint64 b = Fetch64(s + len - 8); |
| 124 | return HashLen16(a, RotateByAtLeast1(b + len, len)) ^ b; |
| 125 | } |
| 126 | if (len >= 4) { |
| 127 | uint64 a = Fetch32(s); |
| 128 | return HashLen16(len + (a << 3), Fetch32(s + len - 4)); |
| 129 | } |
| 130 | if (len > 0) { |
| 131 | uint8 a = s[0]; |
| 132 | uint8 b = s[len >> 1]; |
| 133 | uint8 c = s[len - 1]; |
| 134 | uint32 y = static_cast<uint32>(a) + (static_cast<uint32>(b) << 8); |
| 135 | uint32 z = len + (static_cast<uint32>(c) << 2); |
| 136 | return ShiftMix(y * k2 ^ z * k3) * k2; |
| 137 | } |
| 138 | return k2; |
| 139 | } |
| 140 | |
| 141 | // This probably works well for 16-byte strings as well, but it may be overkill |
| 142 | // in that case. |
| 143 | static uint64 HashLen17to32(const char *s, size_t len) { |
| 144 | uint64 a = Fetch64(s) * k1; |
| 145 | uint64 b = Fetch64(s + 8); |
| 146 | uint64 c = Fetch64(s + len - 8) * k2; |
| 147 | uint64 d = Fetch64(s + len - 16) * k0; |
| 148 | return HashLen16(Rotate(a - b, 43) + Rotate(c, 30) + d, |
| 149 | a + Rotate(b ^ k3, 20) - c + len); |
| 150 | } |
| 151 | |
| 152 | // Return a 16-byte hash for 48 bytes. Quick and dirty. |
| 153 | // Callers do best to use "random-looking" values for a and b. |
| 154 | static pair<uint64, uint64> WeakHashLen32WithSeeds( |
| 155 | uint64 w, uint64 x, uint64 y, uint64 z, uint64 a, uint64 b) { |
| 156 | a += w; |
| 157 | b = Rotate(b + a + z, 21); |
| 158 | uint64 c = a; |
| 159 | a += x; |
| 160 | a += y; |
| 161 | b += Rotate(a, 44); |
| 162 | return make_pair(a + z, b + c); |
| 163 | } |
| 164 | |
| 165 | // Return a 16-byte hash for s[0] ... s[31], a, and b. Quick and dirty. |
| 166 | static pair<uint64, uint64> WeakHashLen32WithSeeds( |
| 167 | const char* s, uint64 a, uint64 b) { |
| 168 | return WeakHashLen32WithSeeds(Fetch64(s), |
| 169 | Fetch64(s + 8), |
| 170 | Fetch64(s + 16), |
| 171 | Fetch64(s + 24), |
| 172 | a, |
| 173 | b); |
| 174 | } |
| 175 | |
| 176 | // Return an 8-byte hash for 33 to 64 bytes. |
| 177 | static uint64 HashLen33to64(const char *s, size_t len) { |
| 178 | uint64 z = Fetch64(s + 24); |
| 179 | uint64 a = Fetch64(s) + (len + Fetch64(s + len - 16)) * k0; |
| 180 | uint64 b = Rotate(a + z, 52); |
| 181 | uint64 c = Rotate(a, 37); |
| 182 | a += Fetch64(s + 8); |
| 183 | c += Rotate(a, 7); |
| 184 | a += Fetch64(s + 16); |
| 185 | uint64 vf = a + z; |
| 186 | uint64 vs = b + Rotate(a, 31) + c; |
| 187 | a = Fetch64(s + 16) + Fetch64(s + len - 32); |
| 188 | z = Fetch64(s + len - 8); |
| 189 | b = Rotate(a + z, 52); |
| 190 | c = Rotate(a, 37); |
| 191 | a += Fetch64(s + len - 24); |
| 192 | c += Rotate(a, 7); |
| 193 | a += Fetch64(s + len - 16); |
| 194 | uint64 wf = a + z; |
| 195 | uint64 ws = b + Rotate(a, 31) + c; |
| 196 | uint64 r = ShiftMix((vf + ws) * k2 + (wf + vs) * k0); |
| 197 | return ShiftMix(r * k0 + vs) * k2; |
| 198 | } |
| 199 | |
| 200 | uint64 CityHash64(const char *s, size_t len) { |
| 201 | if (len <= 32) { |
| 202 | if (len <= 16) { |
| 203 | return HashLen0to16(s, len); |
| 204 | } else { |
| 205 | return HashLen17to32(s, len); |
| 206 | } |
| 207 | } else if (len <= 64) { |
| 208 | return HashLen33to64(s, len); |
| 209 | } |
| 210 | |
| 211 | // For strings over 64 bytes we hash the end first, and then as we |
| 212 | // loop we keep 56 bytes of state: v, w, x, y, and z. |
| 213 | uint64 x = Fetch64(s + len - 40); |
| 214 | uint64 y = Fetch64(s + len - 16) + Fetch64(s + len - 56); |
| 215 | uint64 z = HashLen16(Fetch64(s + len - 48) + len, Fetch64(s + len - 24)); |
| 216 | pair<uint64, uint64> v = WeakHashLen32WithSeeds(s + len - 64, len, z); |
| 217 | pair<uint64, uint64> w = WeakHashLen32WithSeeds(s + len - 32, y + k1, x); |
| 218 | x = x * k1 + Fetch64(s); |
| 219 | |
| 220 | // Decrease len to the nearest multiple of 64, and operate on 64-byte chunks. |
| 221 | len = (len - 1) & ~static_cast<size_t>(63); |
| 222 | do { |
| 223 | x = Rotate(x + y + v.first + Fetch64(s + 8), 37) * k1; |
| 224 | y = Rotate(y + v.second + Fetch64(s + 48), 42) * k1; |
| 225 | x ^= w.second; |
| 226 | y += v.first + Fetch64(s + 40); |
| 227 | z = Rotate(z + w.first, 33) * k1; |
| 228 | v = WeakHashLen32WithSeeds(s, v.second * k1, x + w.first); |
| 229 | w = WeakHashLen32WithSeeds(s + 32, z + w.second, y + Fetch64(s + 16)); |
| 230 | std::swap(z, x); |
| 231 | s += 64; |
| 232 | len -= 64; |
| 233 | } while (len != 0); |
| 234 | return HashLen16(HashLen16(v.first, w.first) + ShiftMix(y) * k1 + z, |
| 235 | HashLen16(v.second, w.second) + x); |
| 236 | } |
| 237 | |
| 238 | uint64 CityHash64WithSeed(const char *s, size_t len, uint64 seed) { |
| 239 | return CityHash64WithSeeds(s, len, k2, seed); |
| 240 | } |
| 241 | |
| 242 | uint64 CityHash64WithSeeds(const char *s, size_t len, |
| 243 | uint64 seed0, uint64 seed1) { |
| 244 | return HashLen16(CityHash64(s, len) - seed0, seed1); |
| 245 | } |
| 246 | |
| 247 | // A subroutine for CityHash128(). Returns a decent 128-bit hash for strings |
| 248 | // of any length representable in signed long. Based on City and Murmur. |
| 249 | static uint128 CityMurmur(const char *s, size_t len, uint128 seed) { |
| 250 | uint64 a = Uint128Low64(seed); |
| 251 | uint64 b = Uint128High64(seed); |
| 252 | uint64 c = 0; |
| 253 | uint64 d = 0; |
| 254 | signed long l = len - 16; |
| 255 | if (l <= 0) { // len <= 16 |
| 256 | a = ShiftMix(a * k1) * k1; |
| 257 | c = b * k1 + HashLen0to16(s, len); |
| 258 | d = ShiftMix(a + (len >= 8 ? Fetch64(s) : c)); |
| 259 | } else { // len > 16 |
| 260 | c = HashLen16(Fetch64(s + len - 8) + k1, a); |
| 261 | d = HashLen16(b + len, c + Fetch64(s + len - 16)); |
| 262 | a += d; |
| 263 | do { |
| 264 | a ^= ShiftMix(Fetch64(s) * k1) * k1; |
| 265 | a *= k1; |
| 266 | b ^= a; |
| 267 | c ^= ShiftMix(Fetch64(s + 8) * k1) * k1; |
| 268 | c *= k1; |
| 269 | d ^= c; |
| 270 | s += 16; |
| 271 | l -= 16; |
| 272 | } while (l > 0); |
| 273 | } |
| 274 | a = HashLen16(a, c); |
| 275 | b = HashLen16(d, b); |
| 276 | return uint128(a ^ b, HashLen16(b, a)); |
| 277 | } |
| 278 | |
| 279 | uint128 CityHash128WithSeed(const char *s, size_t len, uint128 seed) { |
| 280 | if (len < 128) { |
| 281 | return CityMurmur(s, len, seed); |
| 282 | } |
| 283 | |
| 284 | // We expect len >= 128 to be the common case. Keep 56 bytes of state: |
| 285 | // v, w, x, y, and z. |
| 286 | pair<uint64, uint64> v, w; |
| 287 | uint64 x = Uint128Low64(seed); |
| 288 | uint64 y = Uint128High64(seed); |
| 289 | uint64 z = len * k1; |
| 290 | v.first = Rotate(y ^ k1, 49) * k1 + Fetch64(s); |
| 291 | v.second = Rotate(v.first, 42) * k1 + Fetch64(s + 8); |
| 292 | w.first = Rotate(y + z, 35) * k1 + x; |
| 293 | w.second = Rotate(x + Fetch64(s + 88), 53) * k1; |
| 294 | |
| 295 | // This is the same inner loop as CityHash64(), manually unrolled. |
| 296 | do { |
| 297 | x = Rotate(x + y + v.first + Fetch64(s + 8), 37) * k1; |
| 298 | y = Rotate(y + v.second + Fetch64(s + 48), 42) * k1; |
| 299 | x ^= w.second; |
| 300 | y += v.first + Fetch64(s + 40); |
| 301 | z = Rotate(z + w.first, 33) * k1; |
| 302 | v = WeakHashLen32WithSeeds(s, v.second * k1, x + w.first); |
| 303 | w = WeakHashLen32WithSeeds(s + 32, z + w.second, y + Fetch64(s + 16)); |
| 304 | std::swap(z, x); |
| 305 | s += 64; |
| 306 | x = Rotate(x + y + v.first + Fetch64(s + 8), 37) * k1; |
| 307 | y = Rotate(y + v.second + Fetch64(s + 48), 42) * k1; |
| 308 | x ^= w.second; |
| 309 | y += v.first + Fetch64(s + 40); |
| 310 | z = Rotate(z + w.first, 33) * k1; |
| 311 | v = WeakHashLen32WithSeeds(s, v.second * k1, x + w.first); |
| 312 | w = WeakHashLen32WithSeeds(s + 32, z + w.second, y + Fetch64(s + 16)); |
| 313 | std::swap(z, x); |
| 314 | s += 64; |
| 315 | len -= 128; |
| 316 | } while (LIKELY(len >= 128)); |
| 317 | x += Rotate(v.first + z, 49) * k0; |
| 318 | z += Rotate(w.first, 37) * k0; |
| 319 | // If 0 < len < 128, hash up to 4 chunks of 32 bytes each from the end of s. |
| 320 | for (size_t tail_done = 0; tail_done < len; ) { |
| 321 | tail_done += 32; |
| 322 | y = Rotate(x + y, 42) * k0 + v.second; |
| 323 | w.first += Fetch64(s + len - tail_done + 16); |
| 324 | x = x * k0 + w.first; |
| 325 | z += w.second + Fetch64(s + len - tail_done); |
| 326 | w.second += v.first; |
| 327 | v = WeakHashLen32WithSeeds(s + len - tail_done, v.first + z, v.second); |
| 328 | } |
| 329 | // At this point our 56 bytes of state should contain more than |
| 330 | // enough information for a strong 128-bit hash. We use two |
| 331 | // different 56-byte-to-8-byte hashes to get a 16-byte final result. |
| 332 | x = HashLen16(x, v.first); |
| 333 | y = HashLen16(y + z, w.first); |
| 334 | return uint128(HashLen16(x + v.second, w.second) + y, |
| 335 | HashLen16(x + w.second, y + v.second)); |
| 336 | } |
| 337 | |
| 338 | uint128 CityHash128(const char *s, size_t len) { |
| 339 | if (len >= 16) { |
| 340 | return CityHash128WithSeed(s + 16, |
| 341 | len - 16, |
| 342 | uint128(Fetch64(s) ^ k3, |
| 343 | Fetch64(s + 8))); |
| 344 | } else if (len >= 8) { |
| 345 | return CityHash128WithSeed(NULL, |
| 346 | 0, |
| 347 | uint128(Fetch64(s) ^ (len * k0), |
| 348 | Fetch64(s + len - 8) ^ k1)); |
| 349 | } else { |
| 350 | return CityHash128WithSeed(s, len, uint128(k0, k1)); |
| 351 | } |
| 352 | } |
| 353 | |
tanjent@gmail.com | 8adb133 | 2013-04-19 19:57:16 +0000 | [diff] [blame] | 354 | #if defined(__SSE4_2__) && defined(__x86_64__) |
tanjent@gmail.com | f3b7897 | 2012-03-01 03:38:55 +0000 | [diff] [blame] | 355 | #include <nmmintrin.h> |
| 356 | |
| 357 | // Requires len >= 240. |
| 358 | static void CityHashCrc256Long(const char *s, size_t len, |
| 359 | uint32 seed, uint64 *result) { |
| 360 | uint64 a = Fetch64(s + 56) + k0; |
| 361 | uint64 b = Fetch64(s + 96) + k0; |
| 362 | uint64 c = result[0] = HashLen16(b, len); |
| 363 | uint64 d = result[1] = Fetch64(s + 120) * k0 + len; |
| 364 | uint64 e = Fetch64(s + 184) + seed; |
| 365 | uint64 f = seed; |
| 366 | uint64 g = 0; |
| 367 | uint64 h = 0; |
| 368 | uint64 i = 0; |
| 369 | uint64 j = 0; |
| 370 | uint64 t = c + d; |
| 371 | |
| 372 | // 240 bytes of input per iter. |
| 373 | size_t iters = len / 240; |
| 374 | len -= iters * 240; |
| 375 | do { |
| 376 | #define CHUNK(multiplier, z) \ |
| 377 | { \ |
| 378 | uint64 old_a = a; \ |
| 379 | a = Rotate(b, 41 ^ z) * multiplier + Fetch64(s); \ |
| 380 | b = Rotate(c, 27 ^ z) * multiplier + Fetch64(s + 8); \ |
| 381 | c = Rotate(d, 41 ^ z) * multiplier + Fetch64(s + 16); \ |
| 382 | d = Rotate(e, 33 ^ z) * multiplier + Fetch64(s + 24); \ |
| 383 | e = Rotate(t, 25 ^ z) * multiplier + Fetch64(s + 32); \ |
| 384 | t = old_a; \ |
| 385 | } \ |
| 386 | f = _mm_crc32_u64(f, a); \ |
| 387 | g = _mm_crc32_u64(g, b); \ |
| 388 | h = _mm_crc32_u64(h, c); \ |
| 389 | i = _mm_crc32_u64(i, d); \ |
| 390 | j = _mm_crc32_u64(j, e); \ |
| 391 | s += 40 |
| 392 | |
| 393 | CHUNK(1, 1); CHUNK(k0, 0); |
| 394 | CHUNK(1, 1); CHUNK(k0, 0); |
| 395 | CHUNK(1, 1); CHUNK(k0, 0); |
| 396 | } while (--iters > 0); |
| 397 | |
| 398 | while (len >= 40) { |
| 399 | CHUNK(k0, 0); |
| 400 | len -= 40; |
| 401 | } |
| 402 | if (len > 0) { |
| 403 | s = s + len - 40; |
| 404 | CHUNK(k0, 0); |
| 405 | } |
| 406 | j += i << 32; |
| 407 | a = HashLen16(a, j); |
| 408 | h += g << 32; |
| 409 | b += h; |
| 410 | c = HashLen16(c, f) + i; |
| 411 | d = HashLen16(d, e + result[0]); |
| 412 | j += e; |
| 413 | i += HashLen16(h, t); |
| 414 | e = HashLen16(a, d) + j; |
| 415 | f = HashLen16(b, c) + a; |
| 416 | g = HashLen16(j, i) + c; |
| 417 | result[0] = e + f + g + h; |
| 418 | a = ShiftMix((a + g) * k0) * k0 + b; |
| 419 | result[1] += a + result[0]; |
| 420 | a = ShiftMix(a * k0) * k0 + c; |
| 421 | result[2] = a + result[1]; |
| 422 | a = ShiftMix((a + e) * k0) * k0; |
| 423 | result[3] = a + result[2]; |
| 424 | } |
| 425 | |
| 426 | // Requires len < 240. |
| 427 | static void CityHashCrc256Short(const char *s, size_t len, uint64 *result) { |
| 428 | char buf[240]; |
| 429 | memcpy(buf, s, len); |
| 430 | memset(buf + len, 0, 240 - len); |
| 431 | CityHashCrc256Long(buf, 240, ~static_cast<uint32>(len), result); |
| 432 | } |
| 433 | |
| 434 | void CityHashCrc256(const char *s, size_t len, uint64 *result) { |
| 435 | if (LIKELY(len >= 240)) { |
| 436 | CityHashCrc256Long(s, len, 0, result); |
| 437 | } else { |
| 438 | CityHashCrc256Short(s, len, result); |
| 439 | } |
| 440 | } |
| 441 | |
| 442 | uint128 CityHashCrc128WithSeed(const char *s, size_t len, uint128 seed) { |
| 443 | if (len <= 900) { |
| 444 | return CityHash128WithSeed(s, len, seed); |
| 445 | } else { |
| 446 | uint64 result[4]; |
| 447 | CityHashCrc256(s, len, result); |
| 448 | uint64 u = Uint128High64(seed) + result[0]; |
| 449 | uint64 v = Uint128Low64(seed) + result[1]; |
| 450 | return uint128(HashLen16(u, v + result[2]), |
| 451 | HashLen16(Rotate(v, 32), u * k0 + result[3])); |
| 452 | } |
| 453 | } |
| 454 | |
| 455 | uint128 CityHashCrc128(const char *s, size_t len) { |
| 456 | if (len <= 900) { |
| 457 | return CityHash128(s, len); |
| 458 | } else { |
| 459 | uint64 result[4]; |
| 460 | CityHashCrc256(s, len, result); |
| 461 | return uint128(result[2], result[3]); |
| 462 | } |
| 463 | } |
| 464 | |
| 465 | #endif |