forked from facebookresearch/faiss
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathrandom.cpp
More file actions
192 lines (141 loc) · 4.78 KB
/
Copy pathrandom.cpp
File metadata and controls
192 lines (141 loc) · 4.78 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
/**
* Copyright (c) Facebook, Inc. and its affiliates.
*
* This source code is licensed under the MIT license found in the
* LICENSE file in the root directory of this source tree.
*/
// -*- c++ -*-
#include <faiss/utils/random.h>
namespace faiss {
/**************************************************
* Random data generation functions
**************************************************/
RandomGenerator::RandomGenerator (int64_t seed)
: mt((unsigned int)seed) {}
int RandomGenerator::rand_int ()
{
return mt() & 0x7fffffff;
}
int64_t RandomGenerator::rand_int64 ()
{
return int64_t(rand_int()) | int64_t(rand_int()) << 31;
}
int RandomGenerator::rand_int (int max)
{
return mt() % max;
}
float RandomGenerator::rand_float ()
{
return mt() / float(mt.max());
}
double RandomGenerator::rand_double ()
{
return mt() / double(mt.max());
}
/***********************************************************************
* Random functions in this C file only exist because Torch
* counterparts are slow and not multi-threaded. Typical use is for
* more than 1-100 billion values. */
/* Generate a set of random floating point values such that x[i] in [0,1]
multi-threading. For this reason, we rely on re-entreant functions. */
void float_rand (float * x, size_t n, int64_t seed)
{
// only try to parallelize on large enough arrays
const size_t nblock = n < 1024 ? 1 : 1024;
RandomGenerator rng0 (seed);
int a0 = rng0.rand_int (), b0 = rng0.rand_int ();
#pragma omp parallel for
for (size_t j = 0; j < nblock; j++) {
RandomGenerator rng (a0 + j * b0);
const size_t istart = j * n / nblock;
const size_t iend = (j + 1) * n / nblock;
for (size_t i = istart; i < iend; i++)
x[i] = rng.rand_float ();
}
}
void float_randn (float * x, size_t n, int64_t seed)
{
// only try to parallelize on large enough arrays
const size_t nblock = n < 1024 ? 1 : 1024;
RandomGenerator rng0 (seed);
int a0 = rng0.rand_int (), b0 = rng0.rand_int ();
#pragma omp parallel for
for (size_t j = 0; j < nblock; j++) {
RandomGenerator rng (a0 + j * b0);
double a = 0, b = 0, s = 0;
int state = 0; /* generate two number per "do-while" loop */
const size_t istart = j * n / nblock;
const size_t iend = (j + 1) * n / nblock;
for (size_t i = istart; i < iend; i++) {
/* Marsaglia's method (see Knuth) */
if (state == 0) {
do {
a = 2.0 * rng.rand_double () - 1;
b = 2.0 * rng.rand_double () - 1;
s = a * a + b * b;
} while (s >= 1.0);
x[i] = a * sqrt(-2.0 * log(s) / s);
}
else
x[i] = b * sqrt(-2.0 * log(s) / s);
state = 1 - state;
}
}
}
/* Integer versions */
void int64_rand (int64_t * x, size_t n, int64_t seed)
{
// only try to parallelize on large enough arrays
const size_t nblock = n < 1024 ? 1 : 1024;
RandomGenerator rng0 (seed);
int a0 = rng0.rand_int (), b0 = rng0.rand_int ();
#pragma omp parallel for
for (size_t j = 0; j < nblock; j++) {
RandomGenerator rng (a0 + j * b0);
const size_t istart = j * n / nblock;
const size_t iend = (j + 1) * n / nblock;
for (size_t i = istart; i < iend; i++)
x[i] = rng.rand_int64 ();
}
}
void int64_rand_max (int64_t * x, size_t n, uint64_t max, int64_t seed)
{
// only try to parallelize on large enough arrays
const size_t nblock = n < 1024 ? 1 : 1024;
RandomGenerator rng0 (seed);
int a0 = rng0.rand_int (), b0 = rng0.rand_int ();
#pragma omp parallel for
for (size_t j = 0; j < nblock; j++) {
RandomGenerator rng (a0 + j * b0);
const size_t istart = j * n / nblock;
const size_t iend = (j + 1) * n / nblock;
for (size_t i = istart; i < iend; i++)
x[i] = rng.rand_int64 () % max;
}
}
void rand_perm (int *perm, size_t n, int64_t seed)
{
for (size_t i = 0; i < n; i++) perm[i] = i;
RandomGenerator rng (seed);
for (size_t i = 0; i + 1 < n; i++) {
int i2 = i + rng.rand_int (n - i);
std::swap(perm[i], perm[i2]);
}
}
void byte_rand (uint8_t * x, size_t n, int64_t seed)
{
// only try to parallelize on large enough arrays
const size_t nblock = n < 1024 ? 1 : 1024;
RandomGenerator rng0 (seed);
int a0 = rng0.rand_int (), b0 = rng0.rand_int ();
#pragma omp parallel for
for (size_t j = 0; j < nblock; j++) {
RandomGenerator rng (a0 + j * b0);
const size_t istart = j * n / nblock;
const size_t iend = (j + 1) * n / nblock;
size_t i;
for (i = istart; i < iend; i++)
x[i] = rng.rand_int64 ();
}
}
} // namespace faiss