line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
|
2
|
|
|
|
|
|
|
/* Our API for random numbers. |
3
|
|
|
|
|
|
|
* |
4
|
|
|
|
|
|
|
* We can use ISAAC, ChaCha20, or something else. |
5
|
|
|
|
|
|
|
* |
6
|
|
|
|
|
|
|
* 3700 ns/word ChaCha20 in Perl |
7
|
|
|
|
|
|
|
* 3100 ns/word Salsa20 in Perl |
8
|
|
|
|
|
|
|
* 1600 ns/word ChaCha8 in Perl |
9
|
|
|
|
|
|
|
* 760 ns/word ISAAC in Perl |
10
|
|
|
|
|
|
|
* |
11
|
|
|
|
|
|
|
* 11.20 ns/word ChaCha20 (openbsd) |
12
|
|
|
|
|
|
|
* 10.31 ns/word ChaCha20 (dj) |
13
|
|
|
|
|
|
|
* 8.66 ns/word ChaCha20 (sse2 Peters) |
14
|
|
|
|
|
|
|
* 6.85 ns/word ChaCha12 (dj) |
15
|
|
|
|
|
|
|
* 5.99 ns/word Tyche |
16
|
|
|
|
|
|
|
* 5.11 ns/word ChaCha8 (dj) |
17
|
|
|
|
|
|
|
* 4.37 ns/word MT19937 (Cokus) |
18
|
|
|
|
|
|
|
* 4.14 ns/word Tyche-i |
19
|
|
|
|
|
|
|
* 3.26 ns/word ISAAC |
20
|
|
|
|
|
|
|
* 3.18 ns/word PCG64 (64-bit state, 64-bit types) |
21
|
|
|
|
|
|
|
* 1.95 ns/word PCG64 (64-bit state, 128-bit types) |
22
|
|
|
|
|
|
|
* 1.84 ns/word ChaCha20 (AVX2 chacha-opt) |
23
|
|
|
|
|
|
|
* 1.48 ns/word Xoroshiro128+ |
24
|
|
|
|
|
|
|
* 1.16 ns/word SplitMix64 |
25
|
|
|
|
|
|
|
* |
26
|
|
|
|
|
|
|
* These functions do locking, the underlying library does not. |
27
|
|
|
|
|
|
|
*/ |
28
|
|
|
|
|
|
|
|
29
|
|
|
|
|
|
|
#include |
30
|
|
|
|
|
|
|
#include |
31
|
|
|
|
|
|
|
#include |
32
|
|
|
|
|
|
|
#include "ptypes.h" |
33
|
|
|
|
|
|
|
#include "csprng.h" |
34
|
|
|
|
|
|
|
|
35
|
|
|
|
|
|
|
#include "chacha.h" |
36
|
|
|
|
|
|
|
#define SEED_BYTES (32+8) |
37
|
|
|
|
|
|
|
#define CSEED(ctx,bytes,data,good) chacha_seed(ctx,bytes,data,good) |
38
|
|
|
|
|
|
|
#define CRBYTES(ctx,bytes,data) chacha_rand_bytes(ctx,bytes,data) |
39
|
|
|
|
|
|
|
#define CIRAND32(ctx) chacha_irand32(ctx) |
40
|
|
|
|
|
|
|
#define CIRAND64(ctx) chacha_irand64(ctx) |
41
|
|
|
|
|
|
|
#define CSELFTEST() chacha_selftest() |
42
|
|
|
|
|
|
|
|
43
|
|
|
|
|
|
|
/* Helper macros, similar to ChaCha, so we're consistent. */ |
44
|
|
|
|
|
|
|
#ifndef U8TO32_LE |
45
|
|
|
|
|
|
|
#define U8TO32_LE(p) \ |
46
|
|
|
|
|
|
|
(((uint32_t)((p)[0]) ) | \ |
47
|
|
|
|
|
|
|
((uint32_t)((p)[1]) << 8) | \ |
48
|
|
|
|
|
|
|
((uint32_t)((p)[2]) << 16) | \ |
49
|
|
|
|
|
|
|
((uint32_t)((p)[3]) << 24)) |
50
|
|
|
|
|
|
|
#endif |
51
|
|
|
|
|
|
|
#define U32TO8_LE(p, v) \ |
52
|
|
|
|
|
|
|
do { \ |
53
|
|
|
|
|
|
|
uint32_t _v = v; \ |
54
|
|
|
|
|
|
|
(p)[0] = (((_v) ) & 0xFFU); \ |
55
|
|
|
|
|
|
|
(p)[1] = (((_v) >> 8) & 0xFFU); \ |
56
|
|
|
|
|
|
|
(p)[2] = (((_v) >> 16) & 0xFFU); \ |
57
|
|
|
|
|
|
|
(p)[3] = (((_v) >> 24) & 0xFFU); \ |
58
|
|
|
|
|
|
|
} while (0) |
59
|
|
|
|
|
|
|
|
60
|
|
|
|
|
|
|
/*****************************************************************************/ |
61
|
|
|
|
|
|
|
|
62
|
|
|
|
|
|
|
/* We put a simple 32-bit non-CS PRNG here to help fill small seeds. */ |
63
|
|
|
|
|
|
|
#if 0 |
64
|
|
|
|
|
|
|
/* XOSHIRO128** 32-bit output, 32-bit types, 128-bit state */ |
65
|
|
|
|
|
|
|
static INLINE uint32_t rotl(const uint32_t x, int k) { |
66
|
|
|
|
|
|
|
return (x << k) | (x >> (32 - k)); |
67
|
|
|
|
|
|
|
} |
68
|
|
|
|
|
|
|
uint32_t prng_next(char* ctx) { |
69
|
|
|
|
|
|
|
uint32_t *s = (uint32_t*) ctx; |
70
|
|
|
|
|
|
|
const uint32_t result_starstar = rotl(s[0] * 5, 7) * 9; |
71
|
|
|
|
|
|
|
const uint32_t t = s[1] << 9; |
72
|
|
|
|
|
|
|
s[2] ^= s[0]; s[3] ^= s[1]; s[1] ^= s[2]; s[0] ^= s[3]; |
73
|
|
|
|
|
|
|
s[2] ^= t; |
74
|
|
|
|
|
|
|
s[3] = rotl(s[3], 11); |
75
|
|
|
|
|
|
|
return result_starstar; |
76
|
|
|
|
|
|
|
} |
77
|
|
|
|
|
|
|
char* prng_new(uint32_t a, uint32_t b, uint32_t c, uint32_t d) { |
78
|
|
|
|
|
|
|
uint32_t *state; |
79
|
|
|
|
|
|
|
New(0, state, 4, uint32_t); |
80
|
|
|
|
|
|
|
state[0] = 1; state[1] = b; state[2] = c; state[3] = d; |
81
|
|
|
|
|
|
|
(void) prng_next((char*)state); |
82
|
|
|
|
|
|
|
state[0] += a; |
83
|
|
|
|
|
|
|
(void) prng_next((char*)state); |
84
|
|
|
|
|
|
|
return (char*) state; |
85
|
|
|
|
|
|
|
} |
86
|
|
|
|
|
|
|
#else |
87
|
|
|
|
|
|
|
/* PCG RXS M XS 32. 32-bit output, 32-bit state and types. */ |
88
|
64
|
|
|
|
|
|
uint32_t prng_next(char* ctx) { |
89
|
64
|
|
|
|
|
|
uint32_t *rng = (uint32_t*) ctx; |
90
|
64
|
|
|
|
|
|
uint32_t word, oldstate = rng[0]; |
91
|
64
|
|
|
|
|
|
rng[0] = rng[0] * 747796405U + rng[1]; |
92
|
64
|
|
|
|
|
|
word = ((oldstate >> ((oldstate >> 28u) + 4u)) ^ oldstate) * 277803737u; |
93
|
64
|
|
|
|
|
|
return (word >> 22u) ^ word; |
94
|
|
|
|
|
|
|
} |
95
|
5
|
|
|
|
|
|
char* prng_new(uint32_t a, uint32_t b, uint32_t c, uint32_t d) { |
96
|
|
|
|
|
|
|
uint32_t *state; |
97
|
5
|
|
|
|
|
|
New(0, state, 2, uint32_t); |
98
|
5
|
|
|
|
|
|
state[0] = 0U; |
99
|
5
|
|
|
|
|
|
state[1] = (b << 1u) | 1u; |
100
|
5
|
|
|
|
|
|
(void) prng_next((char*)state); |
101
|
5
|
|
|
|
|
|
state[0] += a; |
102
|
5
|
|
|
|
|
|
(void) prng_next((char*)state); |
103
|
5
|
|
|
|
|
|
state[0] ^= c; |
104
|
5
|
|
|
|
|
|
(void) prng_next((char*)state); |
105
|
5
|
|
|
|
|
|
state[0] ^= d; |
106
|
5
|
|
|
|
|
|
(void) prng_next((char*)state); |
107
|
5
|
|
|
|
|
|
return (char*) state; |
108
|
|
|
|
|
|
|
} |
109
|
|
|
|
|
|
|
#endif |
110
|
|
|
|
|
|
|
|
111
|
|
|
|
|
|
|
/*****************************************************************************/ |
112
|
|
|
|
|
|
|
|
113
|
144
|
|
|
|
|
|
uint32_t csprng_context_size(void) |
114
|
|
|
|
|
|
|
{ |
115
|
144
|
|
|
|
|
|
return sizeof(chacha_context_t); |
116
|
|
|
|
|
|
|
} |
117
|
|
|
|
|
|
|
static char _has_selftest_run = 0; |
118
|
|
|
|
|
|
|
|
119
|
83
|
|
|
|
|
|
void csprng_seed(void *ctx, uint32_t bytes, const unsigned char* data) |
120
|
|
|
|
|
|
|
{ |
121
|
|
|
|
|
|
|
unsigned char seed[SEED_BYTES + 4]; |
122
|
|
|
|
|
|
|
|
123
|
|
|
|
|
|
|
/* If given a short seed, minimize zeros in state */ |
124
|
83
|
100
|
|
|
|
|
if (bytes >= SEED_BYTES) { |
125
|
78
|
|
|
|
|
|
memcpy(seed, data, SEED_BYTES); |
126
|
|
|
|
|
|
|
} else { |
127
|
|
|
|
|
|
|
char* rng; |
128
|
|
|
|
|
|
|
uint32_t a, b, c, d, i; |
129
|
5
|
|
|
|
|
|
memcpy(seed, data, bytes); |
130
|
5
|
|
|
|
|
|
memset(seed+bytes, 0, sizeof(seed)-bytes); |
131
|
5
|
|
|
|
|
|
a = U8TO32_LE((seed + 0)); |
132
|
5
|
|
|
|
|
|
b = U8TO32_LE((seed + 4)); |
133
|
5
|
|
|
|
|
|
c = U8TO32_LE((seed + 8)); |
134
|
5
|
|
|
|
|
|
d = U8TO32_LE((seed + 12)); |
135
|
5
|
|
|
|
|
|
rng = prng_new(a,b,c,d); |
136
|
49
|
100
|
|
|
|
|
for (i = 4*((bytes+3)/4); i < SEED_BYTES; i += 4) |
137
|
44
|
|
|
|
|
|
U32TO8_LE(seed + i, prng_next(rng)); |
138
|
5
|
|
|
|
|
|
Safefree(rng); |
139
|
|
|
|
|
|
|
#if 0 |
140
|
|
|
|
|
|
|
printf("got %u bytes in expanded to %u\n", bytes, SEED_BYTES); |
141
|
|
|
|
|
|
|
printf("from: ");for(i=0;i
|
142
|
|
|
|
|
|
|
printf("to: ");for(i=0;i
|
143
|
|
|
|
|
|
|
#endif |
144
|
|
|
|
|
|
|
} |
145
|
|
|
|
|
|
|
|
146
|
83
|
100
|
|
|
|
|
if (!_has_selftest_run) { |
147
|
72
|
|
|
|
|
|
_has_selftest_run = 1; |
148
|
72
|
|
|
|
|
|
CSELFTEST(); |
149
|
|
|
|
|
|
|
} |
150
|
83
|
|
|
|
|
|
CSEED(ctx, SEED_BYTES, seed, (bytes >= 16)); |
151
|
83
|
|
|
|
|
|
} |
152
|
|
|
|
|
|
|
|
153
|
5
|
|
|
|
|
|
extern void csprng_srand(void* ctx, UV insecure_seed) |
154
|
|
|
|
|
|
|
{ |
155
|
|
|
|
|
|
|
#if BITS_PER_WORD == 32 |
156
|
|
|
|
|
|
|
unsigned char seed[4] = {0}; |
157
|
|
|
|
|
|
|
U32TO8_LE(seed, insecure_seed); |
158
|
|
|
|
|
|
|
csprng_seed(ctx, 4, seed); |
159
|
|
|
|
|
|
|
#else |
160
|
5
|
|
|
|
|
|
unsigned char seed[8] = {0}; |
161
|
5
|
100
|
|
|
|
|
if (insecure_seed <= UVCONST(4294967295)) { |
162
|
4
|
|
|
|
|
|
U32TO8_LE(seed, insecure_seed); |
163
|
4
|
|
|
|
|
|
csprng_seed(ctx, 4, seed); |
164
|
|
|
|
|
|
|
} else { |
165
|
1
|
|
|
|
|
|
U32TO8_LE(seed, insecure_seed); |
166
|
1
|
|
|
|
|
|
U32TO8_LE(seed + 4, (insecure_seed >> 32)); |
167
|
1
|
|
|
|
|
|
csprng_seed(ctx, 8, seed); |
168
|
|
|
|
|
|
|
} |
169
|
|
|
|
|
|
|
#endif |
170
|
5
|
|
|
|
|
|
} |
171
|
|
|
|
|
|
|
|
172
|
56
|
|
|
|
|
|
void csprng_rand_bytes(void* ctx, uint32_t bytes, unsigned char* data) |
173
|
|
|
|
|
|
|
{ |
174
|
56
|
|
|
|
|
|
CRBYTES(ctx, bytes, data); |
175
|
56
|
|
|
|
|
|
} |
176
|
|
|
|
|
|
|
|
177
|
101085
|
|
|
|
|
|
uint32_t irand32(void* ctx) |
178
|
|
|
|
|
|
|
{ |
179
|
101085
|
|
|
|
|
|
return CIRAND32(ctx); |
180
|
|
|
|
|
|
|
} |
181
|
2120
|
|
|
|
|
|
UV irand64(void* ctx) |
182
|
|
|
|
|
|
|
{ |
183
|
|
|
|
|
|
|
#if BITS_PER_WORD < 64 |
184
|
|
|
|
|
|
|
croak("irand64 too many bits for UV"); |
185
|
|
|
|
|
|
|
#else |
186
|
2120
|
|
|
|
|
|
return CIRAND64(ctx); |
187
|
|
|
|
|
|
|
#endif |
188
|
|
|
|
|
|
|
} |
189
|
|
|
|
|
|
|
|
190
|
|
|
|
|
|
|
|
191
|
|
|
|
|
|
|
/*****************************************************************************/ |
192
|
|
|
|
|
|
|
|
193
|
1
|
|
|
|
|
|
int is_csprng_well_seeded(void *ctx) |
194
|
|
|
|
|
|
|
{ |
195
|
1
|
|
|
|
|
|
chacha_context_t *cs = ctx; |
196
|
1
|
|
|
|
|
|
return cs->goodseed; |
197
|
|
|
|
|
|
|
} |
198
|
|
|
|
|
|
|
|
199
|
|
|
|
|
|
|
/* There are many ways to get floats from integers. A few good, many bad. |
200
|
|
|
|
|
|
|
* |
201
|
|
|
|
|
|
|
* Vigna recommends (x64 >> 11) * (1.0 / (1ULL<<53)). |
202
|
|
|
|
|
|
|
* http://xoroshiro.di.unimi.it |
203
|
|
|
|
|
|
|
* Also see alternatives discussed: |
204
|
|
|
|
|
|
|
* http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/VERSIONS/C-LANG/speed-up-real.html |
205
|
|
|
|
|
|
|
* |
206
|
|
|
|
|
|
|
* Melissa O'Neill notes the problem is harder than it looks, doesn't address. |
207
|
|
|
|
|
|
|
* http://www.pcg-random.org/pdf/toms-oneill-pcg-family-v1.02.pdf |
208
|
|
|
|
|
|
|
* |
209
|
|
|
|
|
|
|
* randomstate for numpy uses separate code for each generator. |
210
|
|
|
|
|
|
|
* With the exception of dSFMT, they each one one of: |
211
|
|
|
|
|
|
|
* (x64 >> 11) * (1 / 9007199254740992.0) |
212
|
|
|
|
|
|
|
* ((x32 >> 5) * 67108864.0 + (y32 >> 6)) / 9007199254740992.0 |
213
|
|
|
|
|
|
|
* where the first one is identical to Vigna. |
214
|
|
|
|
|
|
|
* |
215
|
|
|
|
|
|
|
* David Jones recommends the minor 32-bit variant: |
216
|
|
|
|
|
|
|
* ((x32 >> 6) * 134217728.0 + (y32 >> 5)) / 9007199254740992.0 |
217
|
|
|
|
|
|
|
* http://www0.cs.ucl.ac.uk/staff/d.jones/GoodPracticeRNG.pdf |
218
|
|
|
|
|
|
|
* |
219
|
|
|
|
|
|
|
* Taylor Campbell discusses this in: |
220
|
|
|
|
|
|
|
* http://mumble.net/~campbell/tmp/random_real.c |
221
|
|
|
|
|
|
|
* He points out that there are two common non-broken choices, |
222
|
|
|
|
|
|
|
* div by 2^-53 or div by 2^-64, and each are slightly flawed in |
223
|
|
|
|
|
|
|
* different ways. He shows a theoretically better method. |
224
|
|
|
|
|
|
|
*/ |
225
|
|
|
|
|
|
|
|
226
|
|
|
|
|
|
|
/* |
227
|
|
|
|
|
|
|
* We prefer the x64 / 2^-64 method. It seems to produce the best results |
228
|
|
|
|
|
|
|
* and is easiest for ensuring we fill up all the bits. |
229
|
|
|
|
|
|
|
* It is similar to what Geoff Kuenning does in MTwist, though he computes |
230
|
|
|
|
|
|
|
* the constants at runtime to ensure a dodgy compiler won't munge them. |
231
|
|
|
|
|
|
|
*/ |
232
|
|
|
|
|
|
|
#define TO_NV_32 2.3283064365386962890625000000000000000E-10L |
233
|
|
|
|
|
|
|
#define TO_NV_64 5.4210108624275221700372640043497085571E-20L |
234
|
|
|
|
|
|
|
#define TO_NV_96 1.2621774483536188886587657044524579675E-29L |
235
|
|
|
|
|
|
|
#define TO_NV_128 2.9387358770557187699218413430556141945E-39L |
236
|
|
|
|
|
|
|
|
237
|
|
|
|
|
|
|
#define DRAND_32_32 (CIRAND32(ctx) * TO_NV_32) |
238
|
|
|
|
|
|
|
#define DRAND_64_32 (((CIRAND32(ctx)>>5) * 67108864.0 + (CIRAND32(ctx)>>6)) / 9007199254740992.0) |
239
|
|
|
|
|
|
|
#define DRAND_64_64 (CIRAND64(ctx) * TO_NV_64) |
240
|
|
|
|
|
|
|
#define DRAND_128_32 (CIRAND32(ctx) * TO_NV_32 + CIRAND32(ctx) * TO_NV_64 + CIRAND32(ctx) * TO_NV_96 + CIRAND32(ctx) * TO_NV_128) |
241
|
|
|
|
|
|
|
#define DRAND_128_64 (CIRAND64(ctx) * TO_NV_64 + CIRAND64(ctx) * TO_NV_128) |
242
|
|
|
|
|
|
|
|
243
|
6037
|
|
|
|
|
|
NV drand64(void* ctx) |
244
|
|
|
|
|
|
|
{ |
245
|
|
|
|
|
|
|
NV r; |
246
|
|
|
|
|
|
|
#if NVMANTBITS <= 32 |
247
|
|
|
|
|
|
|
r = DRAND_32_32; |
248
|
|
|
|
|
|
|
#elif NVMANTBITS <= 64 |
249
|
6037
|
|
|
|
|
|
r = (BITS_PER_WORD <= 32) ? DRAND_64_32 : DRAND_64_64; |
250
|
|
|
|
|
|
|
#else |
251
|
|
|
|
|
|
|
r = (BITS_PER_WORD <= 32) ? DRAND_128_32 : DRAND_128_64; |
252
|
|
|
|
|
|
|
#endif |
253
|
6037
|
|
|
|
|
|
return r; |
254
|
|
|
|
|
|
|
} |
255
|
|
|
|
|
|
|
|
256
|
|
|
|
|
|
|
/* Return rand 32-bit integer between 0 to n-1 inclusive */ |
257
|
131040
|
|
|
|
|
|
uint32_t urandomm32(void *ctx, uint32_t n) |
258
|
|
|
|
|
|
|
{ |
259
|
|
|
|
|
|
|
uint32_t r, rmin; |
260
|
|
|
|
|
|
|
|
261
|
131040
|
100
|
|
|
|
|
if (n <= 1) |
262
|
1010
|
|
|
|
|
|
return 0; |
263
|
|
|
|
|
|
|
|
264
|
130030
|
|
|
|
|
|
rmin = -n % n; |
265
|
|
|
|
|
|
|
while (1) { |
266
|
130042
|
|
|
|
|
|
r = CIRAND32(ctx); |
267
|
130042
|
100
|
|
|
|
|
if (r >= rmin) |
268
|
130030
|
|
|
|
|
|
break; |
269
|
12
|
|
|
|
|
|
} |
270
|
130030
|
|
|
|
|
|
return r % n; |
271
|
|
|
|
|
|
|
} |
272
|
|
|
|
|
|
|
|
273
|
131233
|
|
|
|
|
|
UV urandomm64(void* ctx, UV n) |
274
|
|
|
|
|
|
|
{ |
275
|
|
|
|
|
|
|
UV r, rmin; |
276
|
|
|
|
|
|
|
|
277
|
131233
|
100
|
|
|
|
|
if (n <= 4294967295UL) |
278
|
130918
|
|
|
|
|
|
return urandomm32(ctx,n); |
279
|
315
|
100
|
|
|
|
|
if (n-1 == 4294967295UL) |
280
|
101
|
|
|
|
|
|
return irand32(ctx); |
281
|
|
|
|
|
|
|
|
282
|
214
|
|
|
|
|
|
rmin = -n % n; |
283
|
|
|
|
|
|
|
while (1) { |
284
|
214
|
|
|
|
|
|
r = CIRAND64(ctx); |
285
|
214
|
50
|
|
|
|
|
if (r >= rmin) |
286
|
214
|
|
|
|
|
|
break; |
287
|
0
|
|
|
|
|
|
} |
288
|
214
|
|
|
|
|
|
return r % n; |
289
|
|
|
|
|
|
|
} |
290
|
|
|
|
|
|
|
|
291
|
3093
|
|
|
|
|
|
UV urandomb(void* ctx, int nbits) |
292
|
|
|
|
|
|
|
{ |
293
|
3093
|
100
|
|
|
|
|
if (nbits == 0) { |
294
|
1
|
|
|
|
|
|
return 0; |
295
|
3092
|
100
|
|
|
|
|
} else if (nbits <= 32) { |
296
|
979
|
|
|
|
|
|
return irand32(ctx) >> (32-nbits); |
297
|
|
|
|
|
|
|
#if BITS_PER_WORD == 64 |
298
|
2113
|
50
|
|
|
|
|
} else if (nbits <= 64) { |
299
|
2113
|
|
|
|
|
|
return irand64(ctx) >> (64-nbits); |
300
|
|
|
|
|
|
|
#endif |
301
|
|
|
|
|
|
|
} |
302
|
0
|
|
|
|
|
|
croak("irand64 too many bits for UV"); |
303
|
|
|
|
|
|
|
} |