2 * Copyright 2016 Facebook, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 * Compute 64-, 96-, and 128-bit Rabin fingerprints, as described in
19 * Michael O. Rabin (1981)
20 * Fingerprinting by Random Polynomials
21 * Center for Research in Computing Technology, Harvard University
22 * Tech Report TR-CSE-03-01
24 * The implementation follows the optimization described in
25 * Andrei Z. Broder (1993)
26 * Some applications of Rabin's fingerprinting method
28 * extended for fingerprints larger than 64 bits, and modified to use
29 * 64-bit instead of 32-bit integers for computation.
31 * The precomputed tables are in FingerprintTable.cpp, which is automatically
32 * generated by ComputeFingerprintTable.cpp.
34 * Benchmarked on 10/13/2009 on a 2.5GHz quad-core Xeon L5420,
35 * - Fingerprint<64>::update64() takes about 12ns
36 * - Fingerprint<96>::update64() takes about 30ns
37 * - Fingerprint<128>::update128() takes about 30ns
38 * (unsurprisingly, Fingerprint<96> and Fingerprint<128> take the
39 * same amount of time, as they both use 128-bit operations; the least
40 * significant 32 bits of Fingerprint<96> will always be 0)
42 * @author Tudor Bosman (tudorb@facebook.com)
49 #include <folly/Range.h>
56 struct FingerprintTable {
57 static const uint64_t poly[1 + (BITS - 1) / 64];
58 static const uint64_t table[8][256][1 + (BITS - 1) / 64];
62 const uint64_t FingerprintTable<BITS>::poly[1 + (BITS - 1) / 64] = {};
65 const uint64_t FingerprintTable<BITS>::table[8][256][1 + (BITS - 1) / 64] = {};
70 * Compute the Rabin fingerprint.
72 * TODO(tudorb): Extend this to allow removing values from the computed
73 * fingerprint (so we can fingerprint a sliding window, as in the Rabin-Karp
74 * string matching algorithm)
76 * update* methods return *this, so you can chain them together:
77 * Fingerprint<96>().update8(x).update(str).update64(val).write(output);
83 // Use a non-zero starting value. We'll use (1 << (BITS-1))
85 for (int i = 1; i < size(); i++)
89 Fingerprint& update8(uint8_t v) {
90 uint8_t out = shlor8(v);
91 xortab(detail::FingerprintTable<BITS>::table[0][out]);
95 // update32 and update64 are convenience functions to update the fingerprint
96 // with 4 and 8 bytes at a time. They are faster than calling update8
97 // in a loop. They process the bytes in big-endian order.
98 Fingerprint& update32(uint32_t v) {
99 uint32_t out = shlor32(v);
100 for (int i = 0; i < 4; i++) {
101 xortab(detail::FingerprintTable<BITS>::table[i][out&0xff]);
107 Fingerprint& update64(uint64_t v) {
108 uint64_t out = shlor64(v);
109 for (int i = 0; i < 8; i++) {
110 xortab(detail::FingerprintTable<BITS>::table[i][out&0xff]);
116 Fingerprint& update(StringPiece str) {
117 // TODO(tudorb): We could be smart and do update64 or update32 if aligned
125 * Return the number of uint64s needed to hold the fingerprint value.
128 return 1 + (BITS-1)/64;
132 * Write the computed fingeprint to an array of size() uint64_t's.
133 * For Fingerprint<64>, size()==1; we write 64 bits in out[0]
134 * For Fingerprint<96>, size()==2; we write 64 bits in out[0] and
135 * the most significant 32 bits of out[1]
136 * For Fingerprint<128>, size()==2; we write 64 bits in out[0] and
139 void write(uint64_t* out) const {
140 for (int i = 0; i < size(); i++) {
146 // XOR the fingerprint with a value from one of the tables.
147 void xortab(const uint64_t* tab) {
148 for (int i = 0; i < size(); i++) {
153 // Helper functions: shift the fingerprint value left by 8/32/64 bits,
154 // return the "out" value (the bits that were shifted out), and add "v"
155 // in the bits on the right.
156 uint8_t shlor8(uint8_t v);
157 uint32_t shlor32(uint32_t v);
158 uint64_t shlor64(uint64_t v);
160 uint64_t fp_[1 + (BITS-1)/64];
163 // Convenience functions
166 * Return the 64-bit Rabin fingerprint of a string.
168 inline uint64_t fingerprint64(StringPiece str) {
170 Fingerprint<64>().update(str).write(&fp);
175 * Compute the 96-bit Rabin fingerprint of a string.
176 * Return the 64 most significant bits in *msb, and the 32 least significant
179 inline void fingerprint96(StringPiece str,
180 uint64_t* msb, uint32_t* lsb) {
182 Fingerprint<96>().update(str).write(fp);
184 *lsb = (uint32_t)(fp[1] >> 32);
188 * Compute the 128-bit Rabin fingerprint of a string.
189 * Return the 64 most significant bits in *msb, and the 64 least significant
192 inline void fingerprint128(StringPiece str,
193 uint64_t* msb, uint64_t* lsb) {
195 Fingerprint<128>().update(str).write(fp);
202 inline uint8_t Fingerprint<64>::shlor8(uint8_t v) {
203 uint8_t out = (uint8_t)(fp_[0] >> 56);
204 fp_[0] = (fp_[0] << 8) | ((uint64_t)v);
209 inline uint32_t Fingerprint<64>::shlor32(uint32_t v) {
210 uint32_t out = (uint32_t)(fp_[0] >> 32);
211 fp_[0] = (fp_[0] << 32) | ((uint64_t)v);
216 inline uint64_t Fingerprint<64>::shlor64(uint64_t v) {
217 uint64_t out = fp_[0];
223 inline uint8_t Fingerprint<96>::shlor8(uint8_t v) {
224 uint8_t out = (uint8_t)(fp_[0] >> 56);
225 fp_[0] = (fp_[0] << 8) | (fp_[1] >> 56);
226 fp_[1] = (fp_[1] << 8) | ((uint64_t)v << 32);
231 inline uint32_t Fingerprint<96>::shlor32(uint32_t v) {
232 uint32_t out = (uint32_t)(fp_[0] >> 32);
233 fp_[0] = (fp_[0] << 32) | (fp_[1] >> 32);
234 fp_[1] = ((uint64_t)v << 32);
239 inline uint64_t Fingerprint<96>::shlor64(uint64_t v) {
240 uint64_t out = fp_[0];
241 fp_[0] = fp_[1] | (v >> 32);
247 inline uint8_t Fingerprint<128>::shlor8(uint8_t v) {
248 uint8_t out = (uint8_t)(fp_[0] >> 56);
249 fp_[0] = (fp_[0] << 8) | (fp_[1] >> 56);
250 fp_[1] = (fp_[1] << 8) | ((uint64_t)v);
255 inline uint32_t Fingerprint<128>::shlor32(uint32_t v) {
256 uint32_t out = (uint32_t)(fp_[0] >> 32);
257 fp_[0] = (fp_[0] << 32) | (fp_[1] >> 32);
258 fp_[1] = (fp_[1] << 32) | ((uint64_t)v);
263 inline uint64_t Fingerprint<128>::shlor64(uint64_t v) {
264 uint64_t out = fp_[0];