Commit | Line | Data |
---|---|---|
e25839d4 JA |
1 | #include <math.h> |
2 | #include <string.h> | |
3 | #include <inttypes.h> | |
4 | #include <stdio.h> | |
5 | #include <unistd.h> | |
6 | #include <sys/types.h> | |
7 | #include <fcntl.h> | |
8 | #include "ieee754.h" | |
9 | #include "../log.h" | |
10 | #include "zipf.h" | |
11 | #include "../minmax.h" | |
ed1860cd | 12 | #include "../hash.h" |
e25839d4 | 13 | |
e6f735f0 | 14 | #define ZIPF_MAX_GEN 10000000UL |
e25839d4 JA |
15 | |
16 | static void zipf_update(struct zipf_state *zs) | |
17 | { | |
4c9060e3 | 18 | unsigned long to_gen; |
e25839d4 JA |
19 | unsigned int i; |
20 | ||
4c9060e3 JA |
21 | /* |
22 | * It can become very costly to generate long sequences. Just cap it at | |
e4839591 JA |
23 | * 10M max, that should be doable in 1-2s on even slow machines. |
24 | * Precision will take a slight hit, but nothing major. | |
4c9060e3 | 25 | */ |
761c2729 | 26 | to_gen = min(zs->nranges, (uint64_t) ZIPF_MAX_GEN); |
e25839d4 | 27 | |
4c9060e3 JA |
28 | for (i = 0; i < to_gen; i++) |
29 | zs->zetan += pow(1.0 / (double) (i + 1), zs->theta); | |
e25839d4 JA |
30 | } |
31 | ||
2316296a JA |
32 | static void shared_rand_init(struct zipf_state *zs, unsigned long nranges, |
33 | unsigned int seed) | |
b2b0b753 JA |
34 | { |
35 | memset(zs, 0, sizeof(*zs)); | |
36 | zs->nranges = nranges; | |
37 | ||
c3546b53 | 38 | init_rand_seed(&zs->rand, seed, 0); |
b2b0b753 JA |
39 | zs->rand_off = __rand(&zs->rand); |
40 | } | |
41 | ||
2316296a JA |
42 | void zipf_init(struct zipf_state *zs, unsigned long nranges, double theta, |
43 | unsigned int seed) | |
e25839d4 | 44 | { |
2316296a | 45 | shared_rand_init(zs, nranges, seed); |
e25839d4 | 46 | |
e25839d4 | 47 | zs->theta = theta; |
1442ba18 | 48 | zs->zeta2 = pow(1.0, zs->theta) + pow(0.5, zs->theta); |
e25839d4 | 49 | |
4c9060e3 | 50 | zipf_update(zs); |
e25839d4 JA |
51 | } |
52 | ||
53 | unsigned long long zipf_next(struct zipf_state *zs) | |
54 | { | |
e25839d4 JA |
55 | double alpha, eta, rand_uni, rand_z; |
56 | unsigned long long n = zs->nranges; | |
57 | unsigned long long val; | |
58 | ||
59 | alpha = 1.0 / (1.0 - zs->theta); | |
60 | eta = (1.0 - pow(2.0 / n, 1.0 - zs->theta)) / (1.0 - zs->zeta2 / zs->zetan); | |
61 | ||
c3546b53 | 62 | rand_uni = (double) __rand(&zs->rand) / (double) FRAND32_MAX; |
e25839d4 JA |
63 | rand_z = rand_uni * zs->zetan; |
64 | ||
65 | if (rand_z < 1.0) | |
66 | val = 1; | |
67 | else if (rand_z < (1.0 + pow(0.5, zs->theta))) | |
68 | val = 2; | |
69 | else | |
70 | val = 1 + (unsigned long long)(n * pow(eta*rand_uni - eta + 1.0, alpha)); | |
71 | ||
8c5e96a4 JA |
72 | val--; |
73 | ||
8348daf9 | 74 | if (!zs->disable_hash) |
8c5e96a4 | 75 | val = __hash_u64(val); |
8348daf9 | 76 | |
8c5e96a4 | 77 | return (val + zs->rand_off) % zs->nranges; |
e25839d4 | 78 | } |
925fee33 | 79 | |
2316296a JA |
80 | void pareto_init(struct zipf_state *zs, unsigned long nranges, double h, |
81 | unsigned int seed) | |
925fee33 | 82 | { |
2316296a | 83 | shared_rand_init(zs, nranges, seed); |
925fee33 | 84 | zs->pareto_pow = log(h) / log(1.0 - h); |
925fee33 JA |
85 | } |
86 | ||
87 | unsigned long long pareto_next(struct zipf_state *zs) | |
88 | { | |
c3546b53 | 89 | double rand = (double) __rand(&zs->rand) / (double) FRAND32_MAX; |
8c5e96a4 JA |
90 | unsigned long long n; |
91 | ||
92 | n = (zs->nranges - 1) * pow(rand, zs->pareto_pow); | |
925fee33 | 93 | |
8348daf9 | 94 | if (!zs->disable_hash) |
8c5e96a4 | 95 | n = __hash_u64(n); |
8348daf9 | 96 | |
8c5e96a4 | 97 | return (n + zs->rand_off) % zs->nranges; |
8348daf9 JA |
98 | } |
99 | ||
100 | void zipf_disable_hash(struct zipf_state *zs) | |
101 | { | |
102 | zs->disable_hash = true; | |
925fee33 | 103 | } |