CommitLineData
79c94bd3
JA
1/*
2 This is a maximally equidistributed combined Tausworthe generator
3 based on code from GNU Scientific Library 1.5 (30 Jun 2004)
4
5 x_n = (s1_n ^ s2_n ^ s3_n)
6
7 s1_{n+1} = (((s1_n & 4294967294) <<12) ^ (((s1_n <<13) ^ s1_n) >>19))
8 s2_{n+1} = (((s2_n & 4294967288) << 4) ^ (((s2_n << 2) ^ s2_n) >>25))
9 s3_{n+1} = (((s3_n & 4294967280) <<17) ^ (((s3_n << 3) ^ s3_n) >>11))
10
11 The period of this generator is about 2^88.
12
13 From: P. L'Ecuyer, "Maximally Equidistributed Combined Tausworthe
14 Generators", Mathematics of Computation, 65, 213 (1996), 203--213.
15
17
18 http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme.ps
19 ftp://ftp.iro.umontreal.ca/pub/simulation/lecuyer/papers/tausme.ps
20
21 There is an erratum in the paper "Tables of Maximally
22 Equidistributed Combined LFSR Generators", Mathematics of
23 Computation, 68, 225 (1999), 261--269:
24 http://www.iro.umontreal.ca/~lecuyer/myftp/papers/tausme2.ps
25
26 ... the k_j most significant bits of z_j must be non-
27 zero, for each j. (Note: this restriction also applies to the
28 computer code given in [4], but was mistakenly not mentioned in
29 that paper.)
30
31 This affects the seeding procedure by imposing the requirement
32 s1 > 1, s2 > 7, s3 > 15.
33
34*/
35
9c42684e 36#include <string.h>
1fbbf72e 37#include "rand.h"
637ef8d9 39#include "../hash.h"
1fbbf72e 40
16dc0710
JA
41int arch_random;
42
c3546b53 43static inline uint64_t __seed(uint64_t x, uint64_t m)
1fbbf72e
JA
44{
45 return (x < m) ? x + m : x;
46}
47
5f2f3569 48static void __init_rand32(struct taus88_state *state, unsigned int seed)
2615cc4b
JA
49{
50 int cranks = 6;
51
52#define LCG(x, seed) ((x) * 69069 ^ (seed))
53
54 state->s1 = __seed(LCG((2^31) + (2^17) + (2^7), seed), 1);
55 state->s2 = __seed(LCG(state->s1, seed), 7);
56 state->s3 = __seed(LCG(state->s2, seed), 15);
57
58 while (cranks--)
c3546b53 59 __rand32(state);
2615cc4b
JA
60}
61
9eff5320 62void __init_rand64(struct taus258_state *state, uint64_t seed)
1fbbf72e 63{
c3546b53
JA
64 int cranks = 6;
65
66#define LCG64(x, seed) ((x) * 6906969069ULL ^ (seed))
67
68 state->s1 = __seed(LCG64((2^31) + (2^17) + (2^7), seed), 1);
69 state->s2 = __seed(LCG64(state->s1, seed), 7);
70 state->s3 = __seed(LCG64(state->s2, seed), 15);
71 state->s4 = __seed(LCG64(state->s3, seed), 33);
72 state->s5 = __seed(LCG64(state->s4, seed), 49);
73
74 while (cranks--)
75 __rand64(state);
2615cc4b
JA
76}
77
de4096e8 78void init_rand(struct frand_state *state, bool use64)
2615cc4b 79{
c3546b53
JA
80 state->use64 = use64;
81
82 if (!use64)
83 __init_rand32(&state->state32, 1);
84 else
85 __init_rand64(&state->state64, 1);
86}
87
eda99d55 88void init_rand_seed(struct frand_state *state, uint64_t seed, bool use64)
c3546b53
JA
89{
90 state->use64 = use64;
91
92 if (!use64)
eda99d55 93 __init_rand32(&state->state32, (unsigned int) seed);
c3546b53
JA
94 else
95 __init_rand64(&state->state64, seed);
1fbbf72e 96}
637ef8d9 97
7ff204c8 98void __fill_random_buf_small(void *buf, unsigned int len, uint64_t seed)
637ef8d9 99{
5366025a
GS
100 uint64_t *b = buf;
101 uint64_t *e = b + len / sizeof(*b);
102 unsigned int rest = len % sizeof(*b);
103
104 for (; b != e; ++b) {
105 *b = seed;
dc4729e3 106 seed = __hash_u64(seed);
637ef8d9 107 }
5366025a
GS
108
109 if (fio_unlikely(rest))
110 __builtin_memcpy(e, &seed, rest);
637ef8d9 111}
7d9fb455 112
7ff204c8
SM
113void __fill_random_buf(void *buf, unsigned int len, uint64_t seed)
114{
9dc528b1
JA
115 static uint64_t prime[] = {1, 2, 3, 5, 7, 11, 13, 17,
116 19, 23, 29, 31, 37, 41, 43, 47};
7ff204c8
SM
117 uint64_t *b, *e, s[CONFIG_SEED_BUCKETS];
118 unsigned int rest;
119 int p;
120
121 /*
122 * Calculate the max index which is multiples of the seed buckets.
123 */
124 rest = (len / sizeof(*b) / CONFIG_SEED_BUCKETS) * CONFIG_SEED_BUCKETS;
125
126 b = buf;
127 e = b + rest;
128
129 rest = len - (rest * sizeof(*b));
130
131 for (p = 0; p < CONFIG_SEED_BUCKETS; p++)
132 s[p] = seed * prime[p];
133
134 for (; b != e; b += CONFIG_SEED_BUCKETS) {
135 for (p = 0; p < CONFIG_SEED_BUCKETS; ++p) {
136 b[p] = s[p];
137 s[p] = __hash_u64(s[p]);
138 }
139 }
140
141 __fill_random_buf_small(b, rest, s[0]);
142}
143
9781c080
MHT
144uint64_t fill_random_buf(struct frand_state *fs, void *buf,
145 unsigned int len)
7d9fb455 146{
0d71aa98 147 uint64_t r = __get_next_seed(fs);
7d9fb455
JA
148
149 __fill_random_buf(buf, len, r);
150 return r;
151}
9c42684e 152
9781c080 153void __fill_random_buf_percentage(uint64_t seed, void *buf,
bc769898
JA
154 unsigned int percentage,
155 unsigned int segment, unsigned int len,
156 char *pattern, unsigned int pbytes)
9c42684e 157{
811ac503 158 unsigned int this_len;
9c42684e 159
811ac503 160 if (percentage == 100) {
d1af2894 161 if (pbytes)
2cac8fcb 162 (void)cpy_pattern(pattern, pbytes, buf, len);
d1af2894
JA
163 else
164 memset(buf, 0, len);
bc769898 165 return;
811ac503
JA
166 }
167
168 if (segment > len)
169 segment = len;
9c42684e 170
9c42684e
JA
171 while (len) {
172 /*
173 * Fill random chunk
174 */
2474b06d 175 this_len = ((unsigned long long)segment * (100 - percentage)) / 100;
9c42684e
JA
176 if (this_len > len)
177 this_len = len;
178
bc769898 179 __fill_random_buf(buf, this_len, seed);
9c42684e
JA
180
181 len -= this_len;
6780906f
JA
182 if (!len)
183 break;
9c42684e 184 buf += this_len;
3363fa44 185 this_len = segment - this_len;
9c42684e 186
811ac503
JA
187 if (this_len > len)
188 this_len = len;
6780906f
JA
189 else if (len - this_len <= sizeof(long))
190 this_len = len;
9c42684e 191
d1af2894 192 if (pbytes)
2cac8fcb 193 (void)cpy_pattern(pattern, pbytes, buf, this_len);
d1af2894
JA
194 else
195 memset(buf, 0, this_len);
bc769898 196
98836539
JA
197 len -= this_len;
198 buf += this_len;
9c42684e 199 }
bc769898
JA
200}
201
9781c080
MHT
202uint64_t fill_random_buf_percentage(struct frand_state *fs, void *buf,
203 unsigned int percentage,
204 unsigned int segment, unsigned int len,
205 char *pattern, unsigned int pbytes)
bc769898 206{
0d71aa98 207 uint64_t r = __get_next_seed(fs);
9c42684e 208
bc769898
JA
209 __fill_random_buf_percentage(r, buf, percentage, segment, len,
210 pattern, pbytes);
9c42684e
JA
211 return r;
212}