1/* Copyright (c) 2016 Vladimir Makarov <vmakarov@gcc.gnu.org> 2 3 Permission is hereby granted, free of charge, to any person 4 obtaining a copy of this software and associated documentation 5 files (the "Software"), to deal in the Software without 6 restriction, including without limitation the rights to use, copy, 7 modify, merge, publish, distribute, sublicense, and/or sell copies 8 of the Software, and to permit persons to whom the Software is 9 furnished to do so, subject to the following conditions: 10 11 The above copyright notice and this permission notice shall be 12 included in all copies or substantial portions of the Software. 13 14 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 15 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 16 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 17 NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 18 BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 19 ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 20 CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 SOFTWARE. 22*/ 23 24/* This file implements MUM (MUltiply and Mix) hashing. We randomize 25 input data by 64x64-bit multiplication and mixing hi- and low-parts 26 of the multiplication result by using an addition and then mix it 27 into the current state. We use prime numbers randomly generated 28 with the equal probability of their bit values for the 29 multiplication. When all primes are used once, the state is 30 randomized and the same prime numbers are used again for data 31 randomization. 32 33 The MUM hashing passes all SMHasher tests. Pseudo Random Number 34 Generator based on MUM also passes NIST Statistical Test Suite for 35 Random and Pseudorandom Number Generators for Cryptographic 36 Applications (version 2.2.1) with 1000 bitstreams each containing 37 1M bits. MUM hashing is also faster Spooky64 and City64 on small 38 strings (at least upto 512-bit) on Haswell and Power7. The MUM bulk 39 speed (speed on very long data) is bigger than Spooky and City on 40 Power7. On Haswell the bulk speed is bigger than Spooky one and 41 close to City speed. */ 42 43#ifndef __MUM_HASH__ 44#define __MUM_HASH__ 45 46#include <stddef.h> 47#include <stdlib.h> 48#include <string.h> 49#include <limits.h> 50 51#ifdef _MSC_VER 52typedef unsigned __int16 uint16_t; 53typedef unsigned __int32 uint32_t; 54typedef unsigned __int64 uint64_t; 55#else 56#include <stdint.h> 57#endif 58 59/* Macro saying to use 128-bit integers implemented by GCC for some 60 targets. */ 61#ifndef _MUM_USE_INT128 62/* In GCC uint128_t is defined if HOST_BITS_PER_WIDE_INT >= 64. 63 HOST_WIDE_INT is long if HOST_BITS_PER_LONG > HOST_BITS_PER_INT, 64 otherwise int. */ 65#if defined(__GNUC__) && UINT_MAX != ULONG_MAX 66#define _MUM_USE_INT128 1 67#else 68#define _MUM_USE_INT128 0 69#endif 70#endif 71 72#if 0 73#if defined(__GNUC__) && ((__GNUC__ == 4) && (__GNUC_MINOR__ >= 9) || (__GNUC__ > 4)) 74#define _MUM_FRESH_GCC 75#endif 76#endif 77 78#if defined(__GNUC__) && !defined(__llvm__) && defined(_MUM_FRESH_GCC) 79#define _MUM_ATTRIBUTE_UNUSED __attribute__((unused)) 80#define _MUM_OPTIMIZE(opts) __attribute__((__optimize__ (opts))) 81#define _MUM_TARGET(opts) __attribute__((__target__ (opts))) 82#else 83#define _MUM_ATTRIBUTE_UNUSED 84#define _MUM_OPTIMIZE(opts) 85#define _MUM_TARGET(opts) 86#endif 87 88 89/* Here are different primes randomly generated with the equal 90 probability of their bit values. They are used to randomize input 91 values. */ 92static uint64_t _mum_hash_step_prime = 0x2e0bb864e9ea7df5ULL; 93static uint64_t _mum_key_step_prime = 0xcdb32970830fcaa1ULL; 94static uint64_t _mum_block_start_prime = 0xc42b5e2e6480b23bULL; 95static uint64_t _mum_unroll_prime = 0x7b51ec3d22f7096fULL; 96static uint64_t _mum_tail_prime = 0xaf47d47c99b1461bULL; 97static uint64_t _mum_finish_prime1 = 0xa9a7ae7ceff79f3fULL; 98static uint64_t _mum_finish_prime2 = 0xaf47d47c99b1461bULL; 99 100static uint64_t _mum_primes [] = { 101 0X9ebdcae10d981691, 0X32b9b9b97a27ac7d, 0X29b5584d83d35bbd, 0X4b04e0e61401255f, 102 0X25e8f7b1f1c9d027, 0X80d4c8c000f3e881, 0Xbd1255431904b9dd, 0X8a3bd4485eee6d81, 103 0X3bc721b2aad05197, 0X71b1a19b907d6e33, 0X525e6c1084a8534b, 0X9e4c2cd340c1299f, 104 0Xde3add92e94caa37, 0X7e14eadb1f65311d, 0X3f5aa40f89812853, 0X33b15a3b587d15c9, 105}; 106 107/* Multiply 64-bit V and P and return sum of high and low parts of the 108 result. */ 109static inline uint64_t 110_mum (uint64_t v, uint64_t p) { 111 uint64_t hi, lo; 112#if _MUM_USE_INT128 113#if defined(__aarch64__) 114 /* AARCH64 needs 2 insns to calculate 128-bit result of the 115 multiplication. If we use a generic code we actually call a 116 function doing 128x128->128 bit multiplication. The function is 117 very slow. */ 118 lo = v * p, hi; 119 asm ("umulh %0, %1, %2" : "=r" (hi) : "r" (v), "r" (p)); 120#else 121 __uint128_t r = (__uint128_t) v * (__uint128_t) p; 122 hi = (uint64_t) (r >> 64); 123 lo = (uint64_t) r; 124#endif 125#else 126 /* Implementation of 64x64->128-bit multiplication by four 32x32->64 127 bit multiplication. */ 128 uint64_t hv = v >> 32, hp = p >> 32; 129 uint64_t lv = (uint32_t) v, lp = (uint32_t) p; 130 uint64_t rh = hv * hp; 131 uint64_t rm_0 = hv * lp; 132 uint64_t rm_1 = hp * lv; 133 uint64_t rl = lv * lp; 134 uint64_t t, carry = 0; 135 136 /* We could ignore a carry bit here if we did not care about the 137 same hash for 32-bit and 64-bit targets. */ 138 t = rl + (rm_0 << 32); 139#ifdef MUM_TARGET_INDEPENDENT_HASH 140 carry = t < rl; 141#endif 142 lo = t + (rm_1 << 32); 143#ifdef MUM_TARGET_INDEPENDENT_HASH 144 carry += lo < t; 145#endif 146 hi = rh + (rm_0 >> 32) + (rm_1 >> 32) + carry; 147#endif 148 /* We could use XOR here too but, for some reasons, on Haswell and 149 Power7 using an addition improves hashing performance by 10% for 150 small strings. */ 151 return hi + lo; 152} 153 154#if defined(_MSC_VER) 155#define _mum_bswap_32(x) _byteswap_uint32_t (x) 156#define _mum_bswap_64(x) _byteswap_uint64_t (x) 157#elif defined(__APPLE__) 158#include <libkern/OSByteOrder.h> 159#define _mum_bswap_32(x) OSSwapInt32 (x) 160#define _mum_bswap_64(x) OSSwapInt64 (x) 161#elif defined(__GNUC__) 162#define _mum_bswap32(x) __builtin_bswap32 (x) 163#define _mum_bswap64(x) __builtin_bswap64 (x) 164#else 165#include <byteswap.h> 166#define _mum_bswap32(x) bswap32 (x) 167#define _mum_bswap64(x) bswap64 (x) 168#endif 169 170static inline uint64_t 171_mum_le (uint64_t v) { 172#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ || !defined(MUM_TARGET_INDEPENDENT_HASH) 173 return v; 174#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 175 return _mum_bswap64 (v); 176#else 177#error "Unknown endianess" 178#endif 179} 180 181static inline uint32_t 182_mum_le32 (uint32_t v) { 183#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ || !defined(MUM_TARGET_INDEPENDENT_HASH) 184 return v; 185#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 186 return _mum_bswap32 (v); 187#else 188#error "Unknown endianess" 189#endif 190} 191 192/* Macro defining how many times the most nested loop in 193 _mum_hash_aligned will be unrolled by the compiler (although it can 194 make an own decision:). Use only a constant here to help a 195 compiler to unroll a major loop. 196 197 The macro value affects the result hash for strings > 128 bit. The 198 unroll factor greatly affects the hashing speed. We prefer the 199 speed. */ 200#ifndef _MUM_UNROLL_FACTOR_POWER 201#if defined(__PPC64__) && !defined(MUM_TARGET_INDEPENDENT_HASH) 202#define _MUM_UNROLL_FACTOR_POWER 3 203#elif defined(__aarch64__) && !defined(MUM_TARGET_INDEPENDENT_HASH) 204#define _MUM_UNROLL_FACTOR_POWER 4 205#else 206#define _MUM_UNROLL_FACTOR_POWER 2 207#endif 208#endif 209 210#if _MUM_UNROLL_FACTOR_POWER < 1 211#error "too small unroll factor" 212#elif _MUM_UNROLL_FACTOR_POWER > 4 213#error "We have not enough primes for such unroll factor" 214#endif 215 216#define _MUM_UNROLL_FACTOR (1 << _MUM_UNROLL_FACTOR_POWER) 217 218static inline uint64_t _MUM_OPTIMIZE("unroll-loops") 219_mum_hash_aligned (uint64_t start, const void *key, size_t len) { 220 uint64_t result = start; 221 const unsigned char *str = (const unsigned char *) key; 222 uint64_t u64; 223 int i; 224 size_t n; 225 226 result = _mum (result, _mum_block_start_prime); 227 while (len > _MUM_UNROLL_FACTOR * sizeof (uint64_t)) { 228 /* This loop could be vectorized when we have vector insns for 229 64x64->128-bit multiplication. AVX2 currently only have a 230 vector insn for 4 32x32->64-bit multiplication. */ 231 for (i = 0; i < _MUM_UNROLL_FACTOR; i++) 232 result ^= _mum (_mum_le (((uint64_t *) str)[i]), _mum_primes[i]); 233 len -= _MUM_UNROLL_FACTOR * sizeof (uint64_t); 234 str += _MUM_UNROLL_FACTOR * sizeof (uint64_t); 235 /* We will use the same prime numbers on the next iterations -- 236 randomize the state. */ 237 result = _mum (result, _mum_unroll_prime); 238 } 239 n = len / sizeof (uint64_t); 240 for (i = 0; i < (int)n; i++) 241 result ^= _mum (_mum_le (((uint64_t *) str)[i]), _mum_primes[i]); 242 len -= n * sizeof (uint64_t); str += n * sizeof (uint64_t); 243 switch (len) { 244 case 7: 245 u64 = _mum_le32 (*(uint32_t *) str); 246 u64 |= (uint64_t) str[4] << 32; 247 u64 |= (uint64_t) str[5] << 40; 248 u64 |= (uint64_t) str[6] << 48; 249 return result ^ _mum (u64, _mum_tail_prime); 250 case 6: 251 u64 = _mum_le32 (*(uint32_t *) str); 252 u64 |= (uint64_t) str[4] << 32; 253 u64 |= (uint64_t) str[5] << 40; 254 return result ^ _mum (u64, _mum_tail_prime); 255 case 5: 256 u64 = _mum_le32 (*(uint32_t *) str); 257 u64 |= (uint64_t) str[4] << 32; 258 return result ^ _mum (u64, _mum_tail_prime); 259 case 4: 260 u64 = _mum_le32 (*(uint32_t *) str); 261 return result ^ _mum (u64, _mum_tail_prime); 262 case 3: 263 u64 = str[0]; 264 u64 |= (uint64_t) str[1] << 8; 265 u64 |= (uint64_t) str[2] << 16; 266 return result ^ _mum (u64, _mum_tail_prime); 267 case 2: 268 u64 = str[0]; 269 u64 |= (uint64_t) str[1] << 8; 270 return result ^ _mum (u64, _mum_tail_prime); 271 case 1: 272 u64 = str[0]; 273 return result ^ _mum (u64, _mum_tail_prime); 274 } 275 return result; 276} 277 278/* Final randomization of H. */ 279static inline uint64_t 280_mum_final (uint64_t h) { 281 h ^= _mum (h, _mum_finish_prime1); 282 h ^= _mum (h, _mum_finish_prime2); 283 return h; 284} 285 286#if defined(__x86_64__) && defined(_MUM_FRESH_GCC) 287 288/* We want to use AVX2 insn MULX instead of generic x86-64 MULQ where 289 it is possible. Although on modern Intel processors MULQ takes 290 3-cycles vs. 4 for MULX, MULX permits more freedom in insn 291 scheduling as it uses less fixed registers. */ 292static inline uint64_t _MUM_TARGET("arch=haswell") 293_mum_hash_avx2 (const void * key, size_t len, uint64_t seed) { 294 return _mum_final (_mum_hash_aligned (seed + len, key, len)); 295} 296#endif 297 298#ifndef _MUM_UNALIGNED_ACCESS 299#if defined(__x86_64__) || defined(__i386__) || defined(__PPC64__) \ 300 || defined(__s390__) || defined(__m32c__) || defined(cris) \ 301 || defined(__CR16__) || defined(__vax__) || defined(__m68k__) \ 302 || defined(__aarch64__) 303#define _MUM_UNALIGNED_ACCESS 1 304#else 305#define _MUM_UNALIGNED_ACCESS 0 306#endif 307#endif 308 309/* When we need an aligned access to data being hashed we move part of 310 the unaligned data to an aligned block of given size and then 311 process it, repeating processing the data by the block. */ 312#ifndef _MUM_BLOCK_LEN 313#define _MUM_BLOCK_LEN 1024 314#endif 315 316#if _MUM_BLOCK_LEN < 8 317#error "too small block length" 318#endif 319 320static inline uint64_t 321#if defined(__x86_64__) 322_MUM_TARGET("inline-all-stringops") 323#endif 324_mum_hash_default (const void *key, size_t len, uint64_t seed) { 325 uint64_t result; 326 const unsigned char *str = (const unsigned char *) key; 327 size_t block_len; 328 uint64_t buf[_MUM_BLOCK_LEN / sizeof (uint64_t)]; 329 330 result = seed + len; 331 if (_MUM_UNALIGNED_ACCESS || ((size_t) str & 0x7) == 0) 332 result = _mum_hash_aligned (result, key, len); 333 else { 334 while (len != 0) { 335 block_len = len < _MUM_BLOCK_LEN ? len : _MUM_BLOCK_LEN; 336 memmove (buf, str, block_len); 337 result = _mum_hash_aligned (result, buf, block_len); 338 len -= block_len; 339 str += block_len; 340 } 341 } 342 return _mum_final (result); 343} 344 345static inline uint64_t 346_mum_next_factor (void) { 347 uint64_t start = 0; 348 int i; 349 350 for (i = 0; i < 8; i++) 351 start = (start << 8) | rand() % 256; 352 return start; 353} 354 355/* ++++++++++++++++++++++++++ Interface functions: +++++++++++++++++++ */ 356 357/* Set random multiplicators depending on SEED. */ 358static inline void 359mum_hash_randomize (uint64_t seed) { 360 int i; 361 362 srand (seed); 363 _mum_hash_step_prime = _mum_next_factor (); 364 _mum_key_step_prime = _mum_next_factor (); 365 _mum_finish_prime1 = _mum_next_factor (); 366 _mum_finish_prime2 = _mum_next_factor (); 367 _mum_block_start_prime = _mum_next_factor (); 368 _mum_unroll_prime = _mum_next_factor (); 369 _mum_tail_prime = _mum_next_factor (); 370 for (i = 0; i < (int)(sizeof (_mum_primes) / sizeof (uint64_t)); i++) 371 _mum_primes[i] = _mum_next_factor (); 372} 373 374/* Start hashing data with SEED. Return the state. */ 375static inline uint64_t 376mum_hash_init (uint64_t seed) { 377 return seed; 378} 379 380/* Process data KEY with the state H and return the updated state. */ 381static inline uint64_t 382mum_hash_step (uint64_t h, uint64_t key) 383{ 384 return _mum (h, _mum_hash_step_prime) ^ _mum (key, _mum_key_step_prime); 385} 386 387/* Return the result of hashing using the current state H. */ 388static inline uint64_t 389mum_hash_finish (uint64_t h) { 390 return _mum_final (h); 391} 392 393/* Fast hashing of KEY with SEED. The hash is always the same for the 394 same key on any target. */ 395static inline size_t 396mum_hash64 (uint64_t key, uint64_t seed) { 397 return mum_hash_finish (mum_hash_step (mum_hash_init (seed), key)); 398} 399 400/* Hash data KEY of length LEN and SEED. The hash depends on the 401 target endianess and the unroll factor. */ 402static inline uint64_t 403mum_hash (const void *key, size_t len, uint64_t seed) { 404#if defined(__x86_64__) && defined(_MUM_FRESH_GCC) 405 static int avx2_support = 0; 406 407 if (avx2_support > 0) 408 return _mum_hash_avx2 (key, len, seed); 409 else if (! avx2_support) { 410 __builtin_cpu_init (); 411 avx2_support = __builtin_cpu_supports ("avx2") ? 1 : -1; 412 if (avx2_support > 0) 413 return _mum_hash_avx2 (key, len, seed); 414 } 415#endif 416 return _mum_hash_default (key, len, seed); 417} 418 419#endif 420