]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | //---------------------------------------------------------------------------// |
2 | // Copyright (c) 2015 Muhammad Junaid Muzammil <mjunaidmuzammil@gmail.com> | |
3 | // | |
4 | // Distributed under the Boost Software License, Version 1.0 | |
5 | // See accompanying file LICENSE_1_0.txt or copy at | |
6 | // http://www.boost.org/LICENSE_1_0.txt | |
7 | // | |
8 | // See http://boostorg.github.com/compute for more information. | |
9 | //---------------------------------------------------------------------------// | |
10 | ||
11 | #ifndef BOOST_COMPUTE_RANDOM_THREEFRY_HPP | |
12 | #define BOOST_COMPUTE_RANDOM_THREEFRY_HPP | |
13 | ||
14 | #include <algorithm> | |
15 | ||
16 | #include <boost/compute/types.hpp> | |
17 | #include <boost/compute/buffer.hpp> | |
18 | #include <boost/compute/kernel.hpp> | |
19 | #include <boost/compute/context.hpp> | |
20 | #include <boost/compute/program.hpp> | |
21 | #include <boost/compute/command_queue.hpp> | |
22 | #include <boost/compute/algorithm/transform.hpp> | |
23 | #include <boost/compute/detail/iterator_range_size.hpp> | |
24 | #include <boost/compute/utility/program_cache.hpp> | |
25 | #include <boost/compute/container/vector.hpp> | |
26 | #include <boost/compute/iterator/discard_iterator.hpp> | |
27 | ||
28 | namespace boost { | |
29 | namespace compute { | |
30 | ||
31 | /// \class threefry_engine | |
32 | /// \brief Threefry pseudorandom number generator. | |
33 | template<class T = uint_> | |
34 | class threefry_engine | |
35 | { | |
36 | public: | |
37 | static const size_t threads = 1024; | |
38 | typedef T result_type; | |
39 | ||
40 | /// Creates a new threefry_engine and seeds it with \p value. | |
41 | explicit threefry_engine(command_queue &queue) | |
42 | : m_context(queue.get_context()) | |
43 | { | |
44 | // setup program | |
45 | load_program(); | |
46 | } | |
47 | ||
48 | /// Creates a new threefry_engine object as a copy of \p other. | |
49 | threefry_engine(const threefry_engine<T> &other) | |
50 | : m_context(other.m_context), | |
51 | m_program(other.m_program) | |
52 | { | |
53 | } | |
54 | ||
55 | /// Copies \p other to \c *this. | |
56 | threefry_engine<T>& operator=(const threefry_engine<T> &other) | |
57 | { | |
58 | if(this != &other){ | |
59 | m_context = other.m_context; | |
60 | m_program = other.m_program; | |
61 | } | |
62 | ||
63 | return *this; | |
64 | } | |
65 | ||
66 | /// Destroys the threefry_engine object. | |
67 | ~threefry_engine() | |
68 | { | |
69 | } | |
70 | ||
71 | private: | |
72 | /// \internal_ | |
73 | void load_program() | |
74 | { | |
75 | boost::shared_ptr<program_cache> cache = | |
76 | program_cache::get_global_cache(m_context); | |
77 | std::string cache_key = | |
78 | std::string("threefry_engine_32x2"); | |
79 | ||
80 | // Copyright 2010-2012, D. E. Shaw Research. | |
81 | // All rights reserved. | |
82 | ||
83 | // Redistribution and use in source and binary forms, with or without | |
84 | // modification, are permitted provided that the following conditions are | |
85 | // met: | |
86 | ||
87 | // * Redistributions of source code must retain the above copyright | |
88 | // notice, this list of conditions, and the following disclaimer. | |
89 | ||
90 | // * Redistributions in binary form must reproduce the above copyright | |
91 | // notice, this list of conditions, and the following disclaimer in the | |
92 | // documentation and/or other materials provided with the distribution. | |
93 | ||
94 | // * Neither the name of D. E. Shaw Research nor the names of its | |
95 | // contributors may be used to endorse or promote products derived from | |
96 | // this software without specific prior written permission. | |
97 | ||
98 | // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS | |
99 | // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT | |
100 | // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR | |
101 | // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT | |
102 | // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | |
103 | // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT | |
104 | // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, | |
105 | // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY | |
106 | // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | |
107 | // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | |
108 | // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
109 | const char source[] = | |
110 | "#define THREEFRY2x32_DEFAULT_ROUNDS 20\n" | |
111 | "#define SKEIN_KS_PARITY_32 0x1BD11BDA\n" | |
112 | ||
113 | "enum r123_enum_threefry32x2 {\n" | |
114 | " R_32x2_0_0=13,\n" | |
115 | " R_32x2_1_0=15,\n" | |
116 | " R_32x2_2_0=26,\n" | |
117 | " R_32x2_3_0= 6,\n" | |
118 | " R_32x2_4_0=17,\n" | |
119 | " R_32x2_5_0=29,\n" | |
120 | " R_32x2_6_0=16,\n" | |
121 | " R_32x2_7_0=24\n" | |
122 | "};\n" | |
123 | ||
124 | "static uint RotL_32(uint x, uint N)\n" | |
125 | "{\n" | |
126 | " return (x << (N & 31)) | (x >> ((32-N) & 31));\n" | |
127 | "}\n" | |
128 | ||
129 | "struct r123array2x32 {\n" | |
130 | " uint v[2];\n" | |
131 | "};\n" | |
132 | "typedef struct r123array2x32 threefry2x32_ctr_t;\n" | |
133 | "typedef struct r123array2x32 threefry2x32_key_t;\n" | |
134 | ||
135 | "threefry2x32_ctr_t threefry2x32_R(unsigned int Nrounds, threefry2x32_ctr_t in, threefry2x32_key_t k)\n" | |
136 | "{\n" | |
137 | " threefry2x32_ctr_t X;\n" | |
138 | " uint ks[3];\n" | |
139 | " uint i; \n" | |
140 | " ks[2] = SKEIN_KS_PARITY_32;\n" | |
141 | " for (i=0;i < 2; i++) {\n" | |
142 | " ks[i] = k.v[i];\n" | |
143 | " X.v[i] = in.v[i];\n" | |
144 | " ks[2] ^= k.v[i];\n" | |
145 | " }\n" | |
146 | " X.v[0] += ks[0]; X.v[1] += ks[1];\n" | |
147 | " if(Nrounds>0){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_0_0); X.v[1] ^= X.v[0]; }\n" | |
148 | " if(Nrounds>1){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_1_0); X.v[1] ^= X.v[0]; }\n" | |
149 | " if(Nrounds>2){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_2_0); X.v[1] ^= X.v[0]; }\n" | |
150 | " if(Nrounds>3){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_3_0); X.v[1] ^= X.v[0]; }\n" | |
151 | " if(Nrounds>3){\n" | |
152 | " X.v[0] += ks[1]; X.v[1] += ks[2];\n" | |
153 | " X.v[1] += 1;\n" | |
154 | " }\n" | |
155 | " if(Nrounds>4){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_4_0); X.v[1] ^= X.v[0]; }\n" | |
156 | " if(Nrounds>5){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_5_0); X.v[1] ^= X.v[0]; }\n" | |
157 | " if(Nrounds>6){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_6_0); X.v[1] ^= X.v[0]; }\n" | |
158 | " if(Nrounds>7){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_7_0); X.v[1] ^= X.v[0]; }\n" | |
159 | " if(Nrounds>7){\n" | |
160 | " X.v[0] += ks[2]; X.v[1] += ks[0];\n" | |
161 | " X.v[1] += 2;\n" | |
162 | " }\n" | |
163 | " if(Nrounds>8){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_0_0); X.v[1] ^= X.v[0]; }\n" | |
164 | " if(Nrounds>9){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_1_0); X.v[1] ^= X.v[0]; }\n" | |
165 | " if(Nrounds>10){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_2_0); X.v[1] ^= X.v[0]; }\n" | |
166 | " if(Nrounds>11){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_3_0); X.v[1] ^= X.v[0]; }\n" | |
167 | " if(Nrounds>11){\n" | |
168 | " X.v[0] += ks[0]; X.v[1] += ks[1];\n" | |
169 | " X.v[1] += 3;\n" | |
170 | " }\n" | |
171 | " if(Nrounds>12){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_4_0); X.v[1] ^= X.v[0]; }\n" | |
172 | " if(Nrounds>13){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_5_0); X.v[1] ^= X.v[0]; }\n" | |
173 | " if(Nrounds>14){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_6_0); X.v[1] ^= X.v[0]; }\n" | |
174 | " if(Nrounds>15){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_7_0); X.v[1] ^= X.v[0]; }\n" | |
175 | " if(Nrounds>15){\n" | |
176 | " X.v[0] += ks[1]; X.v[1] += ks[2];\n" | |
177 | " X.v[1] += 4;\n" | |
178 | " }\n" | |
179 | " if(Nrounds>16){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_0_0); X.v[1] ^= X.v[0]; }\n" | |
180 | " if(Nrounds>17){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_1_0); X.v[1] ^= X.v[0]; }\n" | |
181 | " if(Nrounds>18){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_2_0); X.v[1] ^= X.v[0]; }\n" | |
182 | " if(Nrounds>19){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_3_0); X.v[1] ^= X.v[0]; }\n" | |
183 | " if(Nrounds>19){\n" | |
184 | " X.v[0] += ks[2]; X.v[1] += ks[0];\n" | |
185 | " X.v[1] += 5;\n" | |
186 | " }\n" | |
187 | " if(Nrounds>20){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_4_0); X.v[1] ^= X.v[0]; }\n" | |
188 | " if(Nrounds>21){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_5_0); X.v[1] ^= X.v[0]; }\n" | |
189 | " if(Nrounds>22){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_6_0); X.v[1] ^= X.v[0]; }\n" | |
190 | " if(Nrounds>23){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_7_0); X.v[1] ^= X.v[0]; }\n" | |
191 | " if(Nrounds>23){\n" | |
192 | " X.v[0] += ks[0]; X.v[1] += ks[1];\n" | |
193 | " X.v[1] += 6;\n" | |
194 | " }\n" | |
195 | " if(Nrounds>24){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_0_0); X.v[1] ^= X.v[0]; }\n" | |
196 | " if(Nrounds>25){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_1_0); X.v[1] ^= X.v[0]; }\n" | |
197 | " if(Nrounds>26){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_2_0); X.v[1] ^= X.v[0]; }\n" | |
198 | " if(Nrounds>27){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_3_0); X.v[1] ^= X.v[0]; }\n" | |
199 | " if(Nrounds>27){\n" | |
200 | " X.v[0] += ks[1]; X.v[1] += ks[2];\n" | |
201 | " X.v[1] += 7;\n" | |
202 | " }\n" | |
203 | " if(Nrounds>28){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_4_0); X.v[1] ^= X.v[0]; }\n" | |
204 | " if(Nrounds>29){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_5_0); X.v[1] ^= X.v[0]; }\n" | |
205 | " if(Nrounds>30){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_6_0); X.v[1] ^= X.v[0]; }\n" | |
206 | " if(Nrounds>31){ X.v[0] += X.v[1]; X.v[1] = RotL_32(X.v[1],R_32x2_7_0); X.v[1] ^= X.v[0]; }\n" | |
207 | " if(Nrounds>31){\n" | |
208 | " X.v[0] += ks[2]; X.v[1] += ks[0];\n" | |
209 | " X.v[1] += 8;\n" | |
210 | " }\n" | |
211 | " return X;\n" | |
212 | "}\n" | |
213 | ||
214 | "__kernel void generate_rng(__global uint *ctr, __global uint *key, const uint offset) {\n" | |
215 | " threefry2x32_ctr_t in;\n" | |
216 | " threefry2x32_key_t k;\n" | |
217 | " const uint i = get_global_id(0);\n" | |
218 | " in.v[0] = ctr[2 * (offset + i)];\n" | |
219 | " in.v[1] = ctr[2 * (offset + i) + 1];\n" | |
220 | " k.v[0] = key[2 * (offset + i)];\n" | |
221 | " k.v[1] = key[2 * (offset + i) + 1];\n" | |
222 | " in = threefry2x32_R(20, in, k);\n" | |
223 | " ctr[2 * (offset + i)] = in.v[0];\n" | |
224 | " ctr[2 * (offset + i) + 1] = in.v[1];\n" | |
225 | "}\n"; | |
226 | ||
227 | m_program = cache->get_or_build(cache_key, std::string(), source, m_context); | |
228 | } | |
229 | ||
230 | public: | |
231 | ||
232 | ||
233 | /// Generates Threefry random numbers using both the counter and key values, and then stores | |
234 | /// them to the range [\p first_ctr, \p last_ctr). | |
235 | template<class OutputIterator> | |
236 | void generate(OutputIterator first_ctr, OutputIterator last_ctr, OutputIterator first_key, OutputIterator last_key, command_queue &queue) { | |
237 | const size_t size_ctr = detail::iterator_range_size(first_ctr, last_ctr); | |
238 | const size_t size_key = detail::iterator_range_size(first_key, last_key); | |
239 | if(!size_ctr || !size_key || (size_ctr != size_key)) { | |
240 | return; | |
241 | } | |
242 | kernel rng_kernel = m_program.create_kernel("generate_rng"); | |
243 | ||
244 | rng_kernel.set_arg(0, first_ctr.get_buffer()); | |
245 | rng_kernel.set_arg(1, first_key.get_buffer()); | |
246 | size_t offset = 0; | |
247 | ||
248 | for(;;){ | |
249 | size_t count = 0; | |
250 | size_t size = size_ctr/2; | |
251 | if(size > threads){ | |
252 | count = (std::min)(static_cast<size_t>(threads), size - offset); | |
253 | } | |
254 | else { | |
255 | count = size; | |
256 | } | |
257 | rng_kernel.set_arg(2, static_cast<const uint_>(offset)); | |
258 | queue.enqueue_1d_range_kernel(rng_kernel, 0, count, 0); | |
259 | ||
260 | offset += count; | |
261 | ||
262 | if(offset >= size){ | |
263 | break; | |
264 | } | |
265 | ||
266 | } | |
267 | } | |
268 | ||
269 | template<class OutputIterator> | |
270 | void generate(OutputIterator first_ctr, OutputIterator last_ctr, command_queue &queue) { | |
271 | const size_t size_ctr = detail::iterator_range_size(first_ctr, last_ctr); | |
272 | if(!size_ctr) { | |
273 | return; | |
274 | } | |
275 | boost::compute::vector<uint_> vector_key(size_ctr, m_context); | |
276 | vector_key.assign(size_ctr, 0, queue); | |
277 | kernel rng_kernel = m_program.create_kernel("generate_rng"); | |
278 | ||
279 | rng_kernel.set_arg(0, first_ctr.get_buffer()); | |
280 | rng_kernel.set_arg(1, vector_key); | |
281 | size_t offset = 0; | |
282 | ||
283 | for(;;){ | |
284 | size_t count = 0; | |
285 | size_t size = size_ctr/2; | |
286 | if(size > threads){ | |
287 | count = (std::min)(static_cast<size_t>(threads), size - offset); | |
288 | } | |
289 | else { | |
290 | count = size; | |
291 | } | |
292 | rng_kernel.set_arg(2, static_cast<const uint_>(offset)); | |
293 | queue.enqueue_1d_range_kernel(rng_kernel, 0, count, 0); | |
294 | ||
295 | offset += count; | |
296 | ||
297 | if(offset >= size){ | |
298 | break; | |
299 | } | |
300 | ||
301 | } | |
302 | } | |
303 | private: | |
304 | context m_context; | |
305 | program m_program; | |
306 | }; | |
307 | ||
308 | } // end compute namespace | |
309 | } // end boost namespace | |
310 | ||
311 | #endif // BOOST_COMPUTE_RANDOM_THREEFRY_HPP |