/root/doris/be/src/gutil/hash/city.cc
Line | Count | Source (jump to first uncovered line) |
1 | | // Copyright 2010 Google Inc. All Rights Reserved. |
2 | | // Authors: gpike@google.com (Geoff Pike), jyrki@google.com (Jyrki Alakuijala) |
3 | | // |
4 | | // This file provides CityHash64() and related functions. |
5 | | // |
6 | | // The externally visible functions follow the naming conventions of |
7 | | // hash.h, where the size of the output is part of the name. For |
8 | | // example, CityHash64 returns a 64-bit hash. The internal helpers do |
9 | | // not have the return type in their name, but instead have names like |
10 | | // HashLenXX or HashLenXXtoYY, where XX and YY are input string lengths. |
11 | | // |
12 | | // Most of the constants and tricks here were copied from murmur.cc or |
13 | | // hash.h, or discovered by trial and error. It's probably possible to further |
14 | | // optimize the code here by writing a program that systematically explores |
15 | | // more of the space of possible hash functions, or by using SIMD instructions. |
16 | | |
17 | | #include "gutil/hash/city.h" |
18 | | |
19 | | // IWYU pragma: no_include <pstl/glue_algorithm_defs.h> |
20 | | |
21 | | #include <sys/types.h> |
22 | | #include <algorithm> |
23 | | #include <iterator> |
24 | | |
25 | | using std::copy; |
26 | | using std::max; |
27 | | using std::min; |
28 | | using std::reverse; |
29 | | using std::sort; |
30 | | using std::swap; |
31 | | #include <utility> |
32 | | |
33 | | using std::make_pair; |
34 | | using std::pair; |
35 | | |
36 | | #include "common/logging.h" |
37 | | |
38 | | #include "gutil/endian.h" |
39 | | #include "gutil/hash/hash128to64.h" |
40 | | #include "gutil/int128.h" |
41 | | #include "gutil/integral_types.h" |
42 | | #include "gutil/port.h" |
43 | | |
44 | | namespace util_hash { |
45 | | |
46 | | // Some primes between 2^63 and 2^64 for various uses. |
47 | | static const uint64 k0 = 0xa5b85c5e198ed849ULL; |
48 | | static const uint64 k1 = 0x8d58ac26afe12e47ULL; |
49 | | static const uint64 k2 = 0xc47b6e9e3a970ed3ULL; |
50 | | static const uint64 k3 = 0xc70f6907e782aa0bULL; |
51 | | |
52 | | // Bitwise right rotate. Normally this will compile to a single |
53 | | // instruction, especially if the shift is a manifest constant. |
54 | 14.5k | static uint64 Rotate(uint64 val, int shift) { |
55 | 14.5k | DCHECK_GE(shift, 0); |
56 | 14.5k | DCHECK_LE(shift, 63); |
57 | | // Avoid shifting by 64: doing so yields an undefined result. |
58 | 14.5k | return shift == 0 ? val : ((val >> shift) | (val << (64 - shift))); |
59 | 14.5k | } |
60 | | |
61 | | // Equivalent to Rotate(), but requires the second arg to be non-zero. |
62 | | // On x86-64, and probably others, it's possible for this to compile |
63 | | // to a single instruction if both args are already in registers. |
64 | 1.84k | static uint64 RotateByAtLeast1(uint64 val, int shift) { |
65 | 1.84k | DCHECK_GE(shift, 1); |
66 | 1.84k | DCHECK_LE(shift, 63); |
67 | 1.84k | return (val >> shift) | (val << (64 - shift)); |
68 | 1.84k | } |
69 | | |
70 | 2.54k | static uint64 ShiftMix(uint64 val) { |
71 | 2.54k | return val ^ (val >> 47); |
72 | 2.54k | } |
73 | | |
74 | 9.46k | static uint64 HashLen16(uint64 u, uint64 v) { |
75 | 9.46k | return Hash128to64(uint128(u, v)); |
76 | 9.46k | } |
77 | | |
78 | 7.13k | static uint64 HashLen0to16(const char* s, size_t len) { |
79 | 7.13k | DCHECK_GE(len, 0); |
80 | 7.13k | DCHECK_LE(len, 16); |
81 | 7.13k | if (len > 8) { |
82 | 1.84k | uint64 a = LittleEndian::Load64(s); |
83 | 1.84k | uint64 b = LittleEndian::Load64(s + len - 8); |
84 | 1.84k | return HashLen16(a, RotateByAtLeast1(b + len, len)) ^ b; |
85 | 1.84k | } |
86 | 5.29k | if (len >= 4) { |
87 | 2.75k | uint64 a = LittleEndian::Load32(s); |
88 | 2.75k | return HashLen16(len + (a << 3), LittleEndian::Load32(s + len - 4)); |
89 | 2.75k | } |
90 | 2.54k | if (len > 0) { |
91 | 2.54k | uint8 a = s[0]; |
92 | 2.54k | uint8 b = s[len >> 1]; |
93 | 2.54k | uint8 c = s[len - 1]; |
94 | 2.54k | uint32 y = static_cast<uint32>(a) + (static_cast<uint32>(b) << 8); |
95 | 2.54k | uint32 z = len + (static_cast<uint32>(c) << 2); |
96 | 2.54k | return ShiftMix(y * k2 ^ z * k3) * k2; |
97 | 2.54k | } |
98 | 0 | return k2; |
99 | 2.54k | } |
100 | | |
101 | | // This probably works well for 16-byte strings as well, but it may be overkill |
102 | | // in that case. |
103 | 4.85k | static uint64 HashLen17to32(const char* s, size_t len) { |
104 | 4.85k | DCHECK_GE(len, 17); |
105 | 4.85k | DCHECK_LE(len, 32); |
106 | 4.85k | uint64 a = LittleEndian::Load64(s) * k1; |
107 | 4.85k | uint64 b = LittleEndian::Load64(s + 8); |
108 | 4.85k | uint64 c = LittleEndian::Load64(s + len - 8) * k2; |
109 | 4.85k | uint64 d = LittleEndian::Load64(s + len - 16) * k0; |
110 | 4.85k | return HashLen16(Rotate(a - b, 43) + Rotate(c, 30) + d, a + Rotate(b ^ k3, 20) - c + len); |
111 | 4.85k | } |
112 | | |
113 | | // Return a 16-byte hash for 48 bytes. Quick and dirty. |
114 | | // Callers do best to use "random-looking" values for a and b. |
115 | | // (For more, see the code review discussion of CL 18799087.) |
116 | | static pair<uint64, uint64> WeakHashLen32WithSeeds(uint64 w, uint64 x, uint64 y, uint64 z, uint64 a, |
117 | 0 | uint64 b) { |
118 | 0 | a += w; |
119 | 0 | b = Rotate(b + a + z, 51); |
120 | 0 | uint64 c = a; |
121 | 0 | a += x; |
122 | 0 | a += y; |
123 | 0 | b += Rotate(a, 23); |
124 | 0 | return make_pair(a + z, b + c); |
125 | 0 | } |
126 | | |
127 | | // Return a 16-byte hash for s[0] ... s[31], a, and b. Quick and dirty. |
128 | 0 | static pair<uint64, uint64> WeakHashLen32WithSeeds(const char* s, uint64 a, uint64 b) { |
129 | 0 | return WeakHashLen32WithSeeds(LittleEndian::Load64(s), LittleEndian::Load64(s + 8), |
130 | 0 | LittleEndian::Load64(s + 16), LittleEndian::Load64(s + 24), a, b); |
131 | 0 | } |
132 | | |
133 | | // Return an 8-byte hash for 33 to 64 bytes. |
134 | 0 | static uint64 HashLen33to64(const char* s, size_t len) { |
135 | 0 | uint64 z = LittleEndian::Load64(s + 24); |
136 | 0 | uint64 a = LittleEndian::Load64(s) + (len + LittleEndian::Load64(s + len - 16)) * k0; |
137 | 0 | uint64 b = Rotate(a + z, 52); |
138 | 0 | uint64 c = Rotate(a, 37); |
139 | 0 | a += LittleEndian::Load64(s + 8); |
140 | 0 | c += Rotate(a, 7); |
141 | 0 | a += LittleEndian::Load64(s + 16); |
142 | 0 | uint64 vf = a + z; |
143 | 0 | uint64 vs = b + Rotate(a, 31) + c; |
144 | 0 | a = LittleEndian::Load64(s + 16) + LittleEndian::Load64(s + len - 32); |
145 | 0 | z += LittleEndian::Load64(s + len - 8); |
146 | 0 | b = Rotate(a + z, 52); |
147 | 0 | c = Rotate(a, 37); |
148 | 0 | a += LittleEndian::Load64(s + len - 24); |
149 | 0 | c += Rotate(a, 7); |
150 | 0 | a += LittleEndian::Load64(s + len - 16); |
151 | 0 | uint64 wf = a + z; |
152 | 0 | uint64 ws = b + Rotate(a, 31) + c; |
153 | 0 | uint64 r = ShiftMix((vf + ws) * k2 + (wf + vs) * k0); |
154 | 0 | return ShiftMix(r * k0 + vs) * k2; |
155 | 0 | } |
156 | | |
157 | 11.9k | uint64 CityHash64(const char* s, size_t len) { |
158 | 11.9k | if (len <= 32) { |
159 | 11.9k | if (len <= 16) { |
160 | 7.13k | return HashLen0to16(s, len); |
161 | 7.13k | } else { |
162 | 4.85k | return HashLen17to32(s, len); |
163 | 4.85k | } |
164 | 11.9k | } else if (len <= 64) { |
165 | 0 | return HashLen33to64(s, len); |
166 | 0 | } |
167 | | |
168 | | // For strings over 64 bytes we hash the end first, and then as we |
169 | | // loop we keep 56 bytes of state: v, w, x, y, and z. |
170 | 0 | uint64 x = LittleEndian::Load64(s + len - 40); |
171 | 0 | uint64 y = LittleEndian::Load64(s + len - 16) + LittleEndian::Load64(s + len - 56); |
172 | 0 | uint64 z = |
173 | 0 | HashLen16(LittleEndian::Load64(s + len - 48) + len, LittleEndian::Load64(s + len - 24)); |
174 | 0 | pair<uint64, uint64> v = WeakHashLen32WithSeeds(s + len - 64, len, z); |
175 | 0 | pair<uint64, uint64> w = WeakHashLen32WithSeeds(s + len - 32, y + k1, x); |
176 | 0 | x = x * k1 + LittleEndian::Load64(s); |
177 | | |
178 | | // Decrease len to the nearest multiple of 64, and operate on 64-byte chunks. |
179 | 0 | len = (len - 1) & ~static_cast<size_t>(63); |
180 | 0 | DCHECK_GT(len, 0); |
181 | 0 | DCHECK_EQ(len, len / 64 * 64); |
182 | 0 | do { |
183 | 0 | x = Rotate(x + y + v.first + LittleEndian::Load64(s + 8), 37) * k1; |
184 | 0 | y = Rotate(y + v.second + LittleEndian::Load64(s + 48), 42) * k1; |
185 | 0 | x ^= w.second; |
186 | 0 | y += v.first + LittleEndian::Load64(s + 40); |
187 | 0 | z = Rotate(z + w.first, 33) * k1; |
188 | 0 | v = WeakHashLen32WithSeeds(s, v.second * k1, x + w.first); |
189 | 0 | w = WeakHashLen32WithSeeds(s + 32, z + w.second, y + LittleEndian::Load64(s + 16)); |
190 | 0 | std::swap(z, x); |
191 | 0 | s += 64; |
192 | 0 | len -= 64; |
193 | 0 | } while (len != 0); |
194 | 0 | return HashLen16(HashLen16(v.first, w.first) + ShiftMix(y) * k1 + z, |
195 | 0 | HashLen16(v.second, w.second) + x); |
196 | 11.9k | } |
197 | | |
198 | 14 | uint64 CityHash64WithSeed(const char* s, size_t len, uint64 seed) { |
199 | 14 | return CityHash64WithSeeds(s, len, k2, seed); |
200 | 14 | } |
201 | | |
202 | 14 | uint64 CityHash64WithSeeds(const char* s, size_t len, uint64 seed0, uint64 seed1) { |
203 | 14 | return HashLen16(CityHash64(s, len) - seed0, seed1); |
204 | 14 | } |
205 | | |
206 | | // A subroutine for CityHash128(). Returns a decent 128-bit hash for strings |
207 | | // of any length representable in ssize_t. Based on City and Murmur128. |
208 | 0 | static uint128 CityMurmur(const char* s, size_t len, uint128 seed) { |
209 | 0 | uint64 a = Uint128Low64(seed); |
210 | 0 | uint64 b = Uint128High64(seed); |
211 | 0 | uint64 c = 0; |
212 | 0 | uint64 d = 0; |
213 | 0 | ssize_t l = len - 16; |
214 | 0 | if (l <= 0) { // len <= 16 |
215 | 0 | c = b * k1 + HashLen0to16(s, len); |
216 | 0 | d = Rotate(a + (len >= 8 ? LittleEndian::Load64(s) : c), 32); |
217 | 0 | } else { // len > 16 |
218 | 0 | c = HashLen16(LittleEndian::Load64(s + len - 8) + k1, a); |
219 | 0 | d = HashLen16(b + len, c + LittleEndian::Load64(s + len - 16)); |
220 | 0 | a += d; |
221 | 0 | do { |
222 | 0 | a ^= ShiftMix(LittleEndian::Load64(s) * k1) * k1; |
223 | 0 | a *= k1; |
224 | 0 | b ^= a; |
225 | 0 | c ^= ShiftMix(LittleEndian::Load64(s + 8) * k1) * k1; |
226 | 0 | c *= k1; |
227 | 0 | d ^= c; |
228 | 0 | s += 16; |
229 | 0 | l -= 16; |
230 | 0 | } while (l > 0); |
231 | 0 | } |
232 | 0 | a = HashLen16(a, c); |
233 | 0 | b = HashLen16(d, b); |
234 | 0 | return uint128(a ^ b, HashLen16(b, a)); |
235 | 0 | } |
236 | | |
237 | 0 | uint128 CityHash128WithSeed(const char* s, size_t len, uint128 seed) { |
238 | | // TODO(user): As of February 2011, there's a beta of Murmur3 that would |
239 | | // most likely be useful here. E.g., if (len < 900) return Murmur3(...) |
240 | 0 | if (len < 128) { |
241 | 0 | return CityMurmur(s, len, seed); |
242 | 0 | } |
243 | | |
244 | | // We expect len >= 128 to be the common case. Keep 56 bytes of state: |
245 | | // v, w, x, y, and z. |
246 | 0 | pair<uint64, uint64> v, w; |
247 | 0 | uint64 x = Uint128Low64(seed); |
248 | 0 | uint64 y = Uint128High64(seed); |
249 | 0 | uint64 z = len * k1; |
250 | 0 | v.first = Rotate(y ^ k1, 49) * k1 + LittleEndian::Load64(s); |
251 | 0 | v.second = Rotate(v.first, 42) * k1 + LittleEndian::Load64(s + 8); |
252 | 0 | w.first = Rotate(y + z, 35) * k1 + x; |
253 | 0 | w.second = Rotate(x + LittleEndian::Load64(s + 88), 53) * k1; |
254 | | |
255 | | // This is similar to the inner loop of CityHash64(), manually unrolled. |
256 | 0 | do { |
257 | 0 | x = Rotate(x + y + v.first + LittleEndian::Load64(s + 16), 37) * k1; |
258 | 0 | y = Rotate(y + v.second + LittleEndian::Load64(s + 48), 42) * k1; |
259 | 0 | x ^= w.second; |
260 | 0 | y ^= v.first; |
261 | 0 | z = Rotate(z ^ w.first, 33); |
262 | 0 | v = WeakHashLen32WithSeeds(s, v.second * k1, x + w.first); |
263 | 0 | w = WeakHashLen32WithSeeds(s + 32, z + w.second, y); |
264 | 0 | std::swap(z, x); |
265 | 0 | s += 64; |
266 | 0 | x = Rotate(x + y + v.first + LittleEndian::Load64(s + 16), 37) * k1; |
267 | 0 | y = Rotate(y + v.second + LittleEndian::Load64(s + 48), 42) * k1; |
268 | 0 | x ^= w.second; |
269 | 0 | y ^= v.first; |
270 | 0 | z = Rotate(z ^ w.first, 33); |
271 | 0 | v = WeakHashLen32WithSeeds(s, v.second * k1, x + w.first); |
272 | 0 | w = WeakHashLen32WithSeeds(s + 32, z + w.second, y); |
273 | 0 | std::swap(z, x); |
274 | 0 | s += 64; |
275 | 0 | len -= 128; |
276 | 0 | } while (PREDICT_TRUE(len >= 128)); |
277 | 0 | y += Rotate(w.first, 37) * k0 + z; |
278 | 0 | x += Rotate(v.first + z, 49) * k0; |
279 | | // If 0 < len < 128, hash up to 4 chunks of 32 bytes each from the end of s. |
280 | 0 | for (size_t tail_done = 0; tail_done < len;) { |
281 | 0 | tail_done += 32; |
282 | 0 | y = Rotate(y - x, 42) * k0 + v.second; |
283 | 0 | w.first += LittleEndian::Load64(s + len - tail_done + 16); |
284 | 0 | x = Rotate(x, 49) * k0 + w.first; |
285 | 0 | w.first += v.first; |
286 | 0 | v = WeakHashLen32WithSeeds(s + len - tail_done, v.first, v.second); |
287 | 0 | } |
288 | | // At this point our 48 bytes of state should contain more than |
289 | | // enough information for a strong 128-bit hash. We use two |
290 | | // different 48-byte-to-8-byte hashes to get a 16-byte final result. |
291 | 0 | x = HashLen16(x, v.first); |
292 | 0 | y = HashLen16(y, w.first); |
293 | 0 | return uint128(HashLen16(x + v.second, w.second) + y, HashLen16(x + w.second, y + v.second)); |
294 | 0 | } |
295 | | |
296 | 0 | uint128 CityHash128(const char* s, size_t len) { |
297 | 0 | if (len >= 16) { |
298 | 0 | return CityHash128WithSeed( |
299 | 0 | s + 16, len - 16, |
300 | 0 | uint128(LittleEndian::Load64(s) ^ k3, LittleEndian::Load64(s + 8))); |
301 | 0 | } else if (len >= 8) { |
302 | 0 | return CityHash128WithSeed(nullptr, 0, |
303 | 0 | uint128(LittleEndian::Load64(s) ^ (len * k0), |
304 | 0 | LittleEndian::Load64(s + len - 8) ^ k1)); |
305 | 0 | } else { |
306 | 0 | return CityHash128WithSeed(s, len, uint128(k0, k1)); |
307 | 0 | } |
308 | 0 | } |
309 | | |
310 | | } // namespace util_hash |