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