|
| 1 | +//////////////////////////////////////////////////////////////////////////////// |
| 2 | +// |
| 3 | +// X256PP.hh : xoshiro256++ (std::uniform_random_bit_generator) |
| 4 | +// |
| 5 | +// Copyright (c) 2024 Brett Hale. |
| 6 | +// SPDX-License-Identifier: BSD-2-Clause |
| 7 | +// |
| 8 | +//////////////////////////////////////////////////////////////////////////////// |
| 9 | + |
| 10 | + |
| 11 | +#ifndef X256PP_HH_ |
| 12 | +#define X256PP_HH_ |
| 13 | + |
| 14 | + |
| 15 | +#include <cstdint> |
| 16 | +#include <array> |
| 17 | + |
| 18 | +#include <random> |
| 19 | + |
| 20 | + |
| 21 | +//////////////////////////////////////////////////////////////////////////////// |
| 22 | +// |
| 23 | +// X256PP class: |
| 24 | + |
| 25 | + |
| 26 | +class X256PP |
| 27 | +{ |
| 28 | +public: |
| 29 | + |
| 30 | + // xoshiro256++ PRNG (satisfies: UniformRandomBitGenerator) |
| 31 | + |
| 32 | + // based on the 'xoshiro256plusplus.c' implementation by |
| 33 | + // David Blackman and Sebastiano Vigna: prng.di.unimi.it |
| 34 | + |
| 35 | + using result_type = std::uint_fast64_t; |
| 36 | + |
| 37 | + static constexpr result_type result_max = |
| 38 | + static_cast<result_type>(UINT64_C(0xffffffffffffffff)); |
| 39 | + |
| 40 | + static constexpr result_type min () { return (0); } |
| 41 | + static constexpr result_type max () { return result_max; } |
| 42 | + |
| 43 | + |
| 44 | + std::array<result_type, 4> state; // (256 bits) |
| 45 | + |
| 46 | + static constexpr auto clamp (result_type u) { |
| 47 | + return (u & result_max); } |
| 48 | + |
| 49 | + static constexpr auto rol (result_type u, unsigned k) { |
| 50 | + u = clamp(u), k %= 64; // idiomatic rotate: |
| 51 | + return (k) ? clamp((u << k) | (u >> (64 - k))) : (u); |
| 52 | + } |
| 53 | + |
| 54 | + // xoshiro256++ generator function: |
| 55 | + |
| 56 | + result_type operator () () noexcept { |
| 57 | + auto r = rol(state[0] + state[3], (23)) + state[0]; |
| 58 | + auto t = clamp(state[1] << (17)); |
| 59 | + state[2] ^= state[0], state[3] ^= state[1]; |
| 60 | + state[1] ^= state[2], state[0] ^= state[3]; |
| 61 | + state[2] ^= t, state[3] = rol(state[3], (45)); |
| 62 | + return clamp(r); |
| 63 | + } |
| 64 | + |
| 65 | + // equivalent to (2^128) calls to generator: |
| 66 | + |
| 67 | + constexpr void jump () noexcept; |
| 68 | + |
| 69 | + // equivalent to (2^192) calls to generator: |
| 70 | + |
| 71 | + constexpr void long_jump () noexcept; |
| 72 | + |
| 73 | + |
| 74 | + // Stafford's Mix13 variant: |
| 75 | + |
| 76 | + static constexpr auto mix (result_type u) { |
| 77 | + u = clamp(u); u ^= (u >> 30); |
| 78 | + u *= UINT64_C(0xbf58476d1ce4e5b9); |
| 79 | + u = clamp(u); u ^= (u >> 27); |
| 80 | + u *= UINT64_C(0x94d049bb133111eb); |
| 81 | + u = clamp(u); u ^= (u >> 31); |
| 82 | + return u; |
| 83 | + } |
| 84 | + |
| 85 | + // (re)initialize the PRNG with a 64 bit seed: |
| 86 | + |
| 87 | + constexpr X256PP (result_type seed = 0) noexcept |
| 88 | + : state {} { init(seed); } |
| 89 | + |
| 90 | + constexpr void init (result_type seed) noexcept { |
| 91 | + for (auto & si : state) // splitmix: |
| 92 | + si = mix(seed += UINT64_C(0x9e3779b97f4a7c15)); |
| 93 | + } |
| 94 | + |
| 95 | + // (re)initialize the PRNG with a 256 bit state, generated |
| 96 | + // using a std::random_device: |
| 97 | + |
| 98 | + X256PP (std::random_device & rdev) |
| 99 | + : state {} { init(rdev); } |
| 100 | + |
| 101 | + void init (std::random_device & rdev) { |
| 102 | + decltype(state) rbuf; |
| 103 | + std::uniform_int_distribution<result_type> d {0, max()}; |
| 104 | + for (auto & ri : rbuf) ri = d(rdev); |
| 105 | + state = rbuf; |
| 106 | + } |
| 107 | +}; |
| 108 | + |
| 109 | + |
| 110 | +//////////////////////////////////////////////////////////////////////////////// |
| 111 | +// |
| 112 | +// X256PP jump function implementations: |
| 113 | + |
| 114 | + |
| 115 | +constexpr auto |
| 116 | +jump (const X256PP & prng) noexcept |
| 117 | +{ |
| 118 | + auto jrng {prng}; jrng.jump(); return jrng; |
| 119 | +} |
| 120 | + |
| 121 | + |
| 122 | +constexpr void |
| 123 | +X256PP::jump () noexcept |
| 124 | +{ |
| 125 | + decltype(state) jbuf {}; |
| 126 | + |
| 127 | + constexpr decltype(state) jtab { |
| 128 | + UINT64_C(0x180ec6d33cfd0aba), UINT64_C(0xd5a61266f0c9392c), |
| 129 | + UINT64_C(0xa9582618e03fc9aa), UINT64_C(0x39abdc4529b1661c) |
| 130 | + }; |
| 131 | + |
| 132 | + for (auto j : jtab) |
| 133 | + { |
| 134 | + for (int b = 0; b < (64); operator () (), b++) |
| 135 | + { |
| 136 | + if (j & (UINT64_C(1) << b)) |
| 137 | + for (unsigned int i = 0; i < jbuf.size(); i++) |
| 138 | + jbuf[i] ^= state[i]; |
| 139 | + } |
| 140 | + } |
| 141 | + |
| 142 | + state = jbuf; |
| 143 | +} |
| 144 | + |
| 145 | + |
| 146 | +constexpr auto |
| 147 | +long_jump (const X256PP & prng) noexcept |
| 148 | +{ |
| 149 | + auto jrng {prng}; jrng.long_jump(); return jrng; |
| 150 | +} |
| 151 | + |
| 152 | + |
| 153 | +constexpr void |
| 154 | +X256PP::long_jump () noexcept |
| 155 | +{ |
| 156 | + decltype(state) jbuf {}; |
| 157 | + |
| 158 | + constexpr decltype(state) jtab { |
| 159 | + UINT64_C(0x76e15d3efefdcbbf), UINT64_C(0xc5004e441c522fb3), |
| 160 | + UINT64_C(0x77710069854ee241), UINT64_C(0x39109bb02acbe635) |
| 161 | + }; |
| 162 | + |
| 163 | + for (auto j : jtab) |
| 164 | + { |
| 165 | + for (int b = 0; b < (64); operator () (), b++) |
| 166 | + { |
| 167 | + if (j & (UINT64_C(1) << b)) |
| 168 | + for (unsigned int i = 0; i < jbuf.size(); i++) |
| 169 | + jbuf[i] ^= state[i]; |
| 170 | + } |
| 171 | + } |
| 172 | + |
| 173 | + state = jbuf; |
| 174 | +} |
| 175 | + |
| 176 | + |
| 177 | +//////////////////////////////////////////////////////////////////////////////// |
| 178 | + |
| 179 | + |
| 180 | +#endif // X256PP_HH_ |
0 commit comments