1 | // Licensed to the Apache Software Foundation (ASF) under one |
2 | // or more contributor license agreements. See the NOTICE file |
3 | // distributed with this work for additional information |
4 | // regarding copyright ownership. The ASF licenses this file |
5 | // to you under the Apache License, Version 2.0 (the |
6 | // "License"); you may not use this file except in compliance |
7 | // with the License. You may obtain a copy of the License at |
8 | // |
9 | // http://www.apache.org/licenses/LICENSE-2.0 |
10 | // |
11 | // Unless required by applicable law or agreed to in writing, |
12 | // software distributed under the License is distributed on an |
13 | // "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY |
14 | // KIND, either express or implied. See the License for the |
15 | // specific language governing permissions and limitations |
16 | // under the License. |
17 | |
18 | //----------------------------------------------------------------------------- |
19 | // MurmurHash3 was written by Austin Appleby, and is placed in the public |
20 | // domain. The author hereby disclaims copyright to this source code. |
21 | |
22 | // Note - The x86 and x64 versions do _not_ produce the same results, as the |
23 | // algorithms are optimized for their respective platforms. You can still |
24 | // compile and run any of them on any platform, but your performance with the |
25 | // non-native version will be less than optimal. |
26 | |
27 | #include "parquet/murmur3.h" |
28 | |
29 | namespace parquet { |
30 | |
31 | #if defined(_MSC_VER) |
32 | |
33 | #define FORCE_INLINE __forceinline |
34 | #define ROTL64(x, y) _rotl64(x, y) |
35 | |
36 | #else // defined(_MSC_VER) |
37 | |
38 | #define FORCE_INLINE inline __attribute__((always_inline)) |
39 | inline uint64_t rotl64(uint64_t x, int8_t r) { return (x << r) | (x >> (64 - r)); } |
40 | #define ROTL64(x, y) rotl64(x, y) |
41 | |
42 | #endif // !defined(_MSC_VER) |
43 | |
44 | #define BIG_CONSTANT(x) (x##LLU) |
45 | |
46 | //----------------------------------------------------------------------------- |
47 | // Block read - if your platform needs to do endian-swapping or can only |
48 | // handle aligned reads, do the conversion here |
49 | |
50 | FORCE_INLINE uint32_t getblock32(const uint32_t* p, int i) { return p[i]; } |
51 | |
52 | FORCE_INLINE uint64_t getblock64(const uint64_t* p, int i) { return p[i]; } |
53 | |
54 | //----------------------------------------------------------------------------- |
55 | // Finalization mix - force all bits of a hash block to avalanche |
56 | |
57 | FORCE_INLINE uint32_t fmix32(uint32_t h) { |
58 | h ^= h >> 16; |
59 | h *= 0x85ebca6b; |
60 | h ^= h >> 13; |
61 | h *= 0xc2b2ae35; |
62 | h ^= h >> 16; |
63 | |
64 | return h; |
65 | } |
66 | |
67 | //---------- |
68 | |
69 | FORCE_INLINE uint64_t fmix64(uint64_t k) { |
70 | k ^= k >> 33; |
71 | k *= BIG_CONSTANT(0xff51afd7ed558ccd); |
72 | k ^= k >> 33; |
73 | k *= BIG_CONSTANT(0xc4ceb9fe1a85ec53); |
74 | k ^= k >> 33; |
75 | |
76 | return k; |
77 | } |
78 | |
79 | //----------------------------------------------------------------------------- |
80 | |
81 | void Hash_x64_128(const void* key, const int len, const uint32_t seed, uint64_t out[2]) { |
82 | const uint8_t* data = (const uint8_t*)key; |
83 | const int nblocks = len / 16; |
84 | |
85 | uint64_t h1 = seed; |
86 | uint64_t h2 = seed; |
87 | |
88 | const uint64_t c1 = BIG_CONSTANT(0x87c37b91114253d5); |
89 | const uint64_t c2 = BIG_CONSTANT(0x4cf5ad432745937f); |
90 | |
91 | //---------- |
92 | // body |
93 | |
94 | const uint64_t* blocks = (const uint64_t*)(data); |
95 | |
96 | for (int i = 0; i < nblocks; i++) { |
97 | uint64_t k1 = getblock64(blocks, i * 2 + 0); |
98 | uint64_t k2 = getblock64(blocks, i * 2 + 1); |
99 | |
100 | k1 *= c1; |
101 | k1 = ROTL64(k1, 31); |
102 | k1 *= c2; |
103 | h1 ^= k1; |
104 | |
105 | h1 = ROTL64(h1, 27); |
106 | h1 += h2; |
107 | h1 = h1 * 5 + 0x52dce729; |
108 | |
109 | k2 *= c2; |
110 | k2 = ROTL64(k2, 33); |
111 | k2 *= c1; |
112 | h2 ^= k2; |
113 | |
114 | h2 = ROTL64(h2, 31); |
115 | h2 += h1; |
116 | h2 = h2 * 5 + 0x38495ab5; |
117 | } |
118 | |
119 | //---------- |
120 | // tail |
121 | |
122 | const uint8_t* tail = (const uint8_t*)(data + nblocks * 16); |
123 | |
124 | uint64_t k1 = 0; |
125 | uint64_t k2 = 0; |
126 | |
127 | switch (len & 15) { |
128 | case 15: |
129 | k2 ^= ((uint64_t)tail[14]) << 48; // fall through |
130 | case 14: |
131 | k2 ^= ((uint64_t)tail[13]) << 40; // fall through |
132 | case 13: |
133 | k2 ^= ((uint64_t)tail[12]) << 32; // fall through |
134 | case 12: |
135 | k2 ^= ((uint64_t)tail[11]) << 24; // fall through |
136 | case 11: |
137 | k2 ^= ((uint64_t)tail[10]) << 16; // fall through |
138 | case 10: |
139 | k2 ^= ((uint64_t)tail[9]) << 8; // fall through |
140 | case 9: |
141 | k2 ^= ((uint64_t)tail[8]) << 0; |
142 | k2 *= c2; |
143 | k2 = ROTL64(k2, 33); |
144 | k2 *= c1; |
145 | h2 ^= k2; // fall through |
146 | |
147 | case 8: |
148 | k1 ^= ((uint64_t)tail[7]) << 56; // fall through |
149 | case 7: |
150 | k1 ^= ((uint64_t)tail[6]) << 48; // fall through |
151 | case 6: |
152 | k1 ^= ((uint64_t)tail[5]) << 40; // fall through |
153 | case 5: |
154 | k1 ^= ((uint64_t)tail[4]) << 32; // fall through |
155 | case 4: |
156 | k1 ^= ((uint64_t)tail[3]) << 24; // fall through |
157 | case 3: |
158 | k1 ^= ((uint64_t)tail[2]) << 16; // fall through |
159 | case 2: |
160 | k1 ^= ((uint64_t)tail[1]) << 8; // fall through |
161 | case 1: |
162 | k1 ^= ((uint64_t)tail[0]) << 0; |
163 | k1 *= c1; |
164 | k1 = ROTL64(k1, 31); |
165 | k1 *= c2; |
166 | h1 ^= k1; |
167 | } |
168 | |
169 | //---------- |
170 | // finalization |
171 | |
172 | h1 ^= len; |
173 | h2 ^= len; |
174 | |
175 | h1 += h2; |
176 | h2 += h1; |
177 | |
178 | h1 = fmix64(h1); |
179 | h2 = fmix64(h2); |
180 | |
181 | h1 += h2; |
182 | h2 += h1; |
183 | |
184 | reinterpret_cast<uint64_t*>(out)[0] = h1; |
185 | reinterpret_cast<uint64_t*>(out)[1] = h2; |
186 | } |
187 | |
188 | template <typename T> |
189 | uint64_t HashHelper(T value, uint32_t seed) { |
190 | uint64_t output[2]; |
191 | Hash_x64_128(reinterpret_cast<void*>(&value), sizeof(T), seed, output); |
192 | return output[0]; |
193 | } |
194 | |
195 | uint64_t MurmurHash3::Hash(int32_t value) const { return HashHelper(value, seed_); } |
196 | |
197 | uint64_t MurmurHash3::Hash(int64_t value) const { return HashHelper(value, seed_); } |
198 | |
199 | uint64_t MurmurHash3::Hash(float value) const { return HashHelper(value, seed_); } |
200 | |
201 | uint64_t MurmurHash3::Hash(double value) const { return HashHelper(value, seed_); } |
202 | |
203 | uint64_t MurmurHash3::Hash(const FLBA* value, uint32_t len) const { |
204 | uint64_t out[2]; |
205 | Hash_x64_128(reinterpret_cast<const void*>(value->ptr), len, seed_, out); |
206 | return out[0]; |
207 | } |
208 | |
209 | uint64_t MurmurHash3::Hash(const Int96* value) const { |
210 | uint64_t out[2]; |
211 | Hash_x64_128(reinterpret_cast<const void*>(value->value), sizeof(value->value), seed_, |
212 | out); |
213 | return out[0]; |
214 | } |
215 | |
216 | uint64_t MurmurHash3::Hash(const ByteArray* value) const { |
217 | uint64_t out[2]; |
218 | Hash_x64_128(reinterpret_cast<const void*>(value->ptr), value->len, seed_, out); |
219 | return out[0]; |
220 | } |
221 | |
222 | } // namespace parquet |
223 | |