lib/PDL/Primitive-xoshiro256plus.c

filter: Statementstmt 44.6% 29 / 65 Branchbran 27.7% 5 / 18 Conditioncond n/a MC/DCmcdc n/a Subroutinesub n/a Podpod n/a totaltotal 40.9% 34 / 83 CC 0 SCAR 0
CC
0
Cov
0%
CRAP 0
« lib/PDL/Ufunc.xs ↑ Summary lib/PDL/Ops.pd »
1 #include <stdint.h>
2 #include <stdlib.h>
3  
4 /* https://prng.di.unimi.it/xoshiro256plus.c, made re-entrant for PDL */
5 /* Written in 2018 by David Blackman and Sebastiano Vigna (vigna@acm.org)
6  
7 To the extent possible under law, the author has dedicated all copyright
8 and related and neighboring rights to this software to the public domain
9 worldwide. This software is distributed without any warranty.
10  
11 See <http://creativecommons.org/publicdomain/zero/1.0/>. */
12  
13 /* This is xoshiro256+ 1.0, our best and fastest generator for floating-point
14 numbers. We suggest to use its upper bits for floating-point
15 generation, as it is slightly faster than xoshiro256++/xoshiro256**. It
16 passes all tests we are aware of except for the lowest three bits,
17 which might fail linearity tests (and just those), so if low linear
18 complexity is not considered an issue (as it is usually the case) it
19 can be used to generate 64-bit outputs, too.
20  
21 We suggest to use a sign test to extract a random Boolean value, and
22 right shifts to extract subsets of bits.
23  
24 The state must be seeded so that it is not everywhere zero. If you have
25 a 64-bit seed, we suggest to seed a splitmix64 generator and use its
26 output to fill s. */
27  
28 10048 static inline uint64_t rotl(const uint64_t x, int k) {
29 10048 return (x << k) | (x >> (64 - k));
30 }
31  
32 /* needs to point at a suitably-initialised 4-long array */
33 10048 uint64_t xoshiro256plus_next(uint64_t *s) {
34 10048 const uint64_t result = s[0] + s[3];
35 10048 const uint64_t t = s[1] << 17;
36 10048 s[2] ^= s[0];
37 10048 s[3] ^= s[1];
38 10048 s[1] ^= s[2];
39 10048 s[0] ^= s[3];
40 10048 s[2] ^= t;
41 10048 s[3] = rotl(s[3], 45);
42 10048 return result;
43 }
44  
45 /* This is the jump function for the generator. It is equivalent
46 to 2^128 calls to next(); it can be used to generate 2^128
47 non-overlapping subsequences for parallel computations. */
48  
49 0 void xoshiro256plus_jump(uint64_t *s) {
50 static const uint64_t JUMP[] = { 0x180ec6d33cfd0aba, 0xd5a61266f0c9392c, 0xa9582618e03fc9aa, 0x39abdc4529b1661c };
51 0 uint64_t s0 = 0;
52 0 uint64_t s1 = 0;
53 0 uint64_t s2 = 0;
54 0 uint64_t s3 = 0;
55 int i, b;
56 0 ▶ for(i = 0; i < sizeof JUMP / sizeof *JUMP; i++)
Branch: 1 branch
BranchTrueFalseTotal
for(i = 0; i < sizeof JUMP / sizeof *JUMP; i++) 0 0 0
57 0 ▶ for(b = 0; b < 64; b++) {
Branch: 1 branch
BranchTrueFalseTotal
for(b = 0; b < 64; b++) { 0 0 0
58 0 ▶ if (JUMP[i] & UINT64_C(1) << b) {
Branch: 1 branch
BranchTrueFalseTotal
if (JUMP[i] & UINT64_C(1) << b) { 0 0 0
59 0 s0 ^= s[0];
60 0 s1 ^= s[1];
61 0 s2 ^= s[2];
62 0 s3 ^= s[3];
63 }
64 0 xoshiro256plus_next(s);
65 }
66 0 s[0] = s0;
67 0 s[1] = s1;
68 0 s[2] = s2;
69 0 s[3] = s3;
70 0 }
71  
72  
73 /* This is the long-jump function for the generator. It is equivalent to
74 2^192 calls to next(); it can be used to generate 2^64 starting points,
75 from each of which jump() will generate 2^64 non-overlapping
76 subsequences for parallel distributed computations. */
77  
78 0 void xoshiro256plus_long_jump(uint64_t *s) {
79 static const uint64_t LONG_JUMP[] = { 0x76e15d3efefdcbbf, 0xc5004e441c522fb3, 0x77710069854ee241, 0x39109bb02acbe635 };
80 0 uint64_t s0 = 0;
81 0 uint64_t s1 = 0;
82 0 uint64_t s2 = 0;
83 0 uint64_t s3 = 0;
84 int i, b;
85 0 ▶ for(i = 0; i < sizeof LONG_JUMP / sizeof *LONG_JUMP; i++)
Branch: 1 branch
BranchTrueFalseTotal
for(i = 0; i < sizeof LONG_JUMP / sizeof *LONG_JUMP; i++) 0 0 0
86 0 ▶ for(b = 0; b < 64; b++) {
Branch: 1 branch
BranchTrueFalseTotal
for(b = 0; b < 64; b++) { 0 0 0
87 0 ▶ if (LONG_JUMP[i] & UINT64_C(1) << b) {
Branch: 1 branch
BranchTrueFalseTotal
if (LONG_JUMP[i] & UINT64_C(1) << b) { 0 0 0
88 0 s0 ^= s[0];
89 0 s1 ^= s[1];
90 0 s2 ^= s[2];
91 0 s3 ^= s[3];
92 }
93 0 xoshiro256plus_next(s);
94 }
95 0 s[0] = s0;
96 0 s[1] = s1;
97 0 s[2] = s2;
98 0 s[3] = s3;
99 0 }
100  
101 /* https://prng.di.unimi.it/splitmix64.c, deleted licence same as above */
102 /* Written in 2015 by Sebastiano Vigna (vigna@acm.org) */
103  
104 /* This is a fixed-increment version of Java 8's SplittableRandom generator
105 See http://dx.doi.org/10.1145/2714064.2660195 and
106 http://docs.oracle.com/javase/8/docs/api/java/util/SplittableRandom.html
107  
108 It is a very fast generator passing BigCrush, and it can be useful if
109 for some reason you absolutely want 64 bits of state. */
110  
111 468 uint64_t splitmix64_next(uint64_t *x) {
112 468 uint64_t z = (*x += 0x9e3779b97f4a7c15);
113 468 z = (z ^ (z >> 30)) * 0xbf58476d1ce4e5b9;
114 468 z = (z ^ (z >> 27)) * 0x94d049bb133111eb;
115 468 return z ^ (z >> 31);
116 }
117  
118 int pdl_srand_threads = -1; /* how many threads initialised for */
119 uint64_t *pdl_rand_state;
120  
121 /* suitably-initialises n 4-long arrays */
122 9 void pdl_srand(uint64_t **sptr, uint64_t seed, int n) {
123 9 uint64_t x = seed, *s = *sptr;
124 9 ▶ if (pdl_srand_threads < n) {
Branch: 1 branch
BranchTrueFalseTotal
if (pdl_srand_threads < n) { 6 3 9
125 6 ▶ if (*sptr) free(*sptr);
Branch: 1 branch
BranchTrueFalseTotal
if (*sptr) free(*sptr); 0 6 6
126 6 *sptr = s = malloc(n * 4 * sizeof(*s));
127 6 pdl_srand_threads = n;
128 }
129 9 n *= 4;
130 int i;
131 477 ▶ for (i = 0; i < n; i++)
Branch: 1 branch
BranchTrueFalseTotal
for (i = 0; i < n; i++) 468 9 477
132 468 s[i] = splitmix64_next(&x);
133 9 }
134  
135 10048 double pdl_drand(uint64_t *s) {
136 /* code from https://prng.di.unimi.it/ */
137 10048 return (xoshiro256plus_next(s) >> 11) * 0x1.0p-53;
138 }
« lib/PDL/Ufunc.xs ↑ Summary lib/PDL/Ops.pd »