/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-rocrand/checkouts/develop/projects/rocrand/library/include/rocrand/rocrand_mrg31k3p.h Source File

/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-rocrand/checkouts/develop/projects/rocrand/library/include/rocrand/rocrand_mrg31k3p.h Source File#

API library: /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-rocrand/checkouts/develop/projects/rocrand/library/include/rocrand/rocrand_mrg31k3p.h Source File
rocrand_mrg31k3p.h
1 // Copyright (c) 2022-2025 Advanced Micro Devices, Inc. All rights reserved.
2 //
3 // Permission is hereby granted, free of charge, to any person obtaining a copy
4 // of this software and associated documentation files (the "Software"), to deal
5 // in the Software without restriction, including without limitation the rights
6 // to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
7 // copies of the Software, and to permit persons to whom the Software is
8 // furnished to do so, subject to the following conditions:
9 //
10 // The above copyright notice and this permission notice shall be included in
11 // all copies or substantial portions of the Software.
12 //
13 // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 // IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 // FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
16 // AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
17 // LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
18 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
19 // THE SOFTWARE.
20 
21 #ifndef ROCRAND_MRG31K3P_H_
22 #define ROCRAND_MRG31K3P_H_
23 
24 #include "rocrand/rocrand_common.h"
25 #include "rocrand/rocrand_mrg31k3p_precomputed.h"
26 
27 #include <hip/hip_runtime.h>
28 
29 #define ROCRAND_MRG31K3P_M1 2147483647U // 2 ^ 31 - 1
30 #define ROCRAND_MRG31K3P_M2 2147462579U // 2 ^ 31 - 21069
31 #define ROCRAND_MRG31K3P_MASK12 511U // 2 ^ 9 - 1
32 #define ROCRAND_MRG31K3P_MASK13 16777215U // 2 ^ 24 - 1
33 #define ROCRAND_MRG31K3P_MASK21 65535U // 2 ^ 16 - 1
34 #define ROCRAND_MRG31K3P_NORM_DOUBLE (4.656612875245796923e-10) // 1 / ROCRAND_MRG31K3P_M1
35 #define ROCRAND_MRG31K3P_UINT32_NORM \
36  (2.000000001396983862) // UINT32_MAX / (ROCRAND_MRG31K3P_M1 - 1)
37 
46 #define ROCRAND_MRG31K3P_DEFAULT_SEED 12345ULL // end of group rocranddevice
48 
49 namespace rocrand_device
50 {
51 
52 class mrg31k3p_engine
53 {
54 public:
55  struct mrg31k3p_state
56  {
57  unsigned int x1[3];
58  unsigned int x2[3];
59 
60 #ifndef ROCRAND_DETAIL_BM_NOT_IN_STATE
61  // The Box–Muller transform requires two inputs to convert uniformly
62  // distributed real values [0; 1] to normally distributed real values
63  // (with mean = 0, and stddev = 1). Often user wants only one
64  // normally distributed number, to save performance and random
65  // numbers the 2nd value is saved for future requests.
66  unsigned int boxmuller_float_state; // is there a float in boxmuller_float
67  unsigned int boxmuller_double_state; // is there a double in boxmuller_double
68  float boxmuller_float; // normally distributed float
69  double boxmuller_double; // normally distributed double
70 #endif
71  };
72 
73  __forceinline__ __device__ __host__ mrg31k3p_engine()
74  {
75  this->seed(ROCRAND_MRG31K3P_DEFAULT_SEED, 0, 0);
76  }
77 
86  __forceinline__ __device__ __host__ mrg31k3p_engine(const unsigned long long seed,
87  const unsigned long long subsequence,
88  const unsigned long long offset)
89  {
90  this->seed(seed, subsequence, offset);
91  }
92 
101  __forceinline__ __device__ __host__ void seed(unsigned long long seed_value,
102  const unsigned long long subsequence,
103  const unsigned long long offset)
104  {
105  if(seed_value == 0)
106  {
107  seed_value = ROCRAND_MRG31K3P_DEFAULT_SEED;
108  }
109  unsigned int x = static_cast<unsigned int>(seed_value ^ 0x55555555U);
110  unsigned int y = static_cast<unsigned int>((seed_value >> 32) ^ 0xAAAAAAAAU);
111  m_state.x1[0] = mod_mul_m1(x, seed_value);
112  m_state.x1[1] = mod_mul_m1(y, seed_value);
113  m_state.x1[2] = mod_mul_m1(x, seed_value);
114  m_state.x2[0] = mod_mul_m2(y, seed_value);
115  m_state.x2[1] = mod_mul_m2(x, seed_value);
116  m_state.x2[2] = mod_mul_m2(y, seed_value);
117  this->restart(subsequence, offset);
118  }
119 
121  __forceinline__ __device__ __host__ void discard(unsigned long long offset)
122  {
123  this->discard_impl(offset);
124  }
125 
128  __forceinline__ __device__ __host__ void discard_subsequence(unsigned long long subsequence)
129  {
130  this->discard_subsequence_impl(subsequence);
131  }
132 
135  __forceinline__ __device__ __host__ void discard_sequence(unsigned long long sequence)
136  {
137  this->discard_sequence_impl(sequence);
138  }
139 
140  __forceinline__ __device__ __host__ void restart(const unsigned long long subsequence,
141  const unsigned long long offset)
142  {
143 #ifndef ROCRAND_DETAIL_BM_NOT_IN_STATE
144  m_state.boxmuller_float_state = 0;
145  m_state.boxmuller_double_state = 0;
146 #endif
147  this->discard_subsequence_impl(subsequence);
148  this->discard_impl(offset);
149  }
150 
151  __forceinline__ __device__ __host__ unsigned int operator()()
152  {
153  return this->next();
154  }
155 
156  // Returned value is in range [1, ROCRAND_MRG31K3P_M1].
157  __forceinline__ __device__ __host__ unsigned int next()
158  {
159  // First component
160  unsigned int tmp
161  = (((m_state.x1[1] & ROCRAND_MRG31K3P_MASK12) << 22) + (m_state.x1[1] >> 9))
162  + (((m_state.x1[2] & ROCRAND_MRG31K3P_MASK13) << 7) + (m_state.x1[2] >> 24));
163  tmp -= (tmp >= ROCRAND_MRG31K3P_M1) ? ROCRAND_MRG31K3P_M1 : 0;
164  tmp += m_state.x1[2];
165  tmp -= (tmp >= ROCRAND_MRG31K3P_M1) ? ROCRAND_MRG31K3P_M1 : 0;
166  m_state.x1[2] = m_state.x1[1];
167  m_state.x1[1] = m_state.x1[0];
168  m_state.x1[0] = tmp;
169 
170  // Second component
171  tmp = (((m_state.x2[0] & ROCRAND_MRG31K3P_MASK21) << 15) + 21069 * (m_state.x2[0] >> 16));
172  tmp -= (tmp >= ROCRAND_MRG31K3P_M2) ? ROCRAND_MRG31K3P_M2 : 0;
173  tmp += ((m_state.x2[2] & ROCRAND_MRG31K3P_MASK21) << 15);
174  tmp -= (tmp >= ROCRAND_MRG31K3P_M2) ? ROCRAND_MRG31K3P_M2 : 0;
175  tmp += 21069 * (m_state.x2[2] >> 16);
176  tmp -= (tmp >= ROCRAND_MRG31K3P_M2) ? ROCRAND_MRG31K3P_M2 : 0;
177  tmp += m_state.x2[2];
178  tmp -= (tmp >= ROCRAND_MRG31K3P_M2) ? ROCRAND_MRG31K3P_M2 : 0;
179  m_state.x2[2] = m_state.x2[1];
180  m_state.x2[1] = m_state.x2[0];
181  m_state.x2[0] = tmp;
182 
183  // Combination
184  return m_state.x1[0] - m_state.x2[0]
185  + (m_state.x1[0] <= m_state.x2[0] ? ROCRAND_MRG31K3P_M1 : 0);
186  }
187 
188 protected:
189  // Advances the internal state to skip \p offset numbers.
190  __forceinline__ __device__ __host__ void discard_impl(unsigned long long offset)
191  {
192  discard_state(offset);
193  }
194 
195  // Advances the internal state to skip \p subsequence subsequences.
196  __forceinline__ __device__ __host__ void
197  discard_subsequence_impl(unsigned long long subsequence)
198  {
199  int i = 0;
200 
201  while(subsequence > 0)
202  {
203  if(subsequence & 1)
204  {
205 #if defined(__HIP_DEVICE_COMPILE__)
206  mod_mat_vec_m1(d_mrg31k3p_A1P72 + i, m_state.x1);
207  mod_mat_vec_m2(d_mrg31k3p_A2P72 + i, m_state.x2);
208 #else
209  mod_mat_vec_m1(h_mrg31k3p_A1P72 + i, m_state.x1);
210  mod_mat_vec_m2(h_mrg31k3p_A2P72 + i, m_state.x2);
211 #endif
212  }
213  subsequence >>= 1;
214  i += 9;
215  }
216  }
217 
218  // Advances the internal state to skip \p sequences.
219  __forceinline__ __device__ __host__ void discard_sequence_impl(unsigned long long sequence)
220  {
221  int i = 0;
222 
223  while(sequence > 0)
224  {
225  if(sequence & 1)
226  {
227 #if defined(__HIP_DEVICE_COMPILE__)
228  mod_mat_vec_m1(d_mrg31k3p_A1P134 + i, m_state.x1);
229  mod_mat_vec_m2(d_mrg31k3p_A2P134 + i, m_state.x2);
230 #else
231  mod_mat_vec_m1(h_mrg31k3p_A1P134 + i, m_state.x1);
232  mod_mat_vec_m2(h_mrg31k3p_A2P134 + i, m_state.x2);
233 #endif
234  }
235  sequence >>= 1;
236  i += 9;
237  }
238  }
239 
240  // Advances the internal state to skip \p offset numbers.
241  __forceinline__ __device__ __host__ void discard_state(unsigned long long offset)
242  {
243  int i = 0;
244 
245  while(offset > 0)
246  {
247  if(offset & 1)
248  {
249 #if defined(__HIP_DEVICE_COMPILE__)
250  mod_mat_vec_m1(d_mrg31k3p_A1 + i, m_state.x1);
251  mod_mat_vec_m2(d_mrg31k3p_A2 + i, m_state.x2);
252 #else
253  mod_mat_vec_m1(h_mrg31k3p_A1 + i, m_state.x1);
254  mod_mat_vec_m2(h_mrg31k3p_A2 + i, m_state.x2);
255 #endif
256  }
257  offset >>= 1;
258  i += 9;
259  }
260  }
261 
262  // Advances the internal state to the next state.
263  __forceinline__ __device__ __host__ void discard_state()
264  {
265  discard_state(1);
266  }
267 
268 private:
269  __forceinline__ __device__ __host__ static void mod_mat_vec_m1(const unsigned int* A,
270  unsigned int* s)
271  {
272  unsigned long long x[3] = {s[0], s[1], s[2]};
273 
274  s[0] = mod_m1(mod_m1(A[0] * x[0]) + mod_m1(A[1] * x[1]) + mod_m1(A[2] * x[2]));
275 
276  s[1] = mod_m1(mod_m1(A[3] * x[0]) + mod_m1(A[4] * x[1]) + mod_m1(A[5] * x[2]));
277 
278  s[2] = mod_m1(mod_m1(A[6] * x[0]) + mod_m1(A[7] * x[1]) + mod_m1(A[8] * x[2]));
279  }
280 
281  __forceinline__ __device__ __host__ static void mod_mat_vec_m2(const unsigned int* A,
282  unsigned int* s)
283  {
284  unsigned long long x[3] = {s[0], s[1], s[2]};
285 
286  s[0] = mod_m2(mod_m2(A[0] * x[0]) + mod_m2(A[1] * x[1]) + mod_m2(A[2] * x[2]));
287 
288  s[1] = mod_m2(mod_m2(A[3] * x[0]) + mod_m2(A[4] * x[1]) + mod_m2(A[5] * x[2]));
289 
290  s[2] = mod_m2(mod_m2(A[6] * x[0]) + mod_m2(A[7] * x[1]) + mod_m2(A[8] * x[2]));
291  }
292 
293  __forceinline__ __device__ __host__ static unsigned long long mod_mul_m1(unsigned int i,
294  unsigned long long j)
295  {
296  return mod_m1(i * j);
297  }
298 
299  __forceinline__ __device__ __host__ static unsigned long long mod_m1(unsigned long long p)
300  {
301  return p % ROCRAND_MRG31K3P_M1;
302  }
303 
304  __forceinline__ __device__ __host__ static unsigned long long mod_mul_m2(unsigned int i,
305  unsigned long long j)
306  {
307  return mod_m2(i * j);
308  }
309 
310  __forceinline__ __device__ __host__ static unsigned long long mod_m2(unsigned long long p)
311  {
312  return p % ROCRAND_MRG31K3P_M2;
313  }
314 
315 protected:
316  // State
317  mrg31k3p_state m_state;
318 
319 #ifndef ROCRAND_DETAIL_BM_NOT_IN_STATE
320  friend struct detail::engine_boxmuller_helper<mrg31k3p_engine>;
321 #endif
322 }; // mrg31k3p_engine class
323 
324 } // end namespace rocrand_device
325 
332 typedef rocrand_device::mrg31k3p_engine rocrand_state_mrg31k3p;
334 
346 __forceinline__ __device__ __host__
347 void rocrand_init(const unsigned long long seed,
348  const unsigned long long subsequence,
349  const unsigned long long offset,
350  rocrand_state_mrg31k3p* state)
351 {
352  *state = rocrand_state_mrg31k3p(seed, subsequence, offset);
353 }
354 
367 __forceinline__ __device__ __host__
368 unsigned int rocrand(rocrand_state_mrg31k3p* state)
369 {
370  // next() in [1, ROCRAND_MRG31K3P_M1]
371  return static_cast<unsigned int>((state->next() - 1) * ROCRAND_MRG31K3P_UINT32_NORM);
372 }
373 
382 __forceinline__ __device__ __host__
383 void skipahead(unsigned long long offset, rocrand_state_mrg31k3p* state)
384 {
385  return state->discard(offset);
386 }
387 
397 __forceinline__ __device__ __host__
398 void skipahead_subsequence(unsigned long long subsequence, rocrand_state_mrg31k3p* state)
399 {
400  return state->discard_subsequence(subsequence);
401 }
402 
412 __forceinline__ __device__ __host__
413 void skipahead_sequence(unsigned long long sequence, rocrand_state_mrg31k3p* state)
414 {
415  return state->discard_sequence(sequence);
416 }
417  // end of group rocranddevice
419 
420 #endif // ROCRAND_MRG31K3P_H_
#define ROCRAND_MRG31K3P_DEFAULT_SEED
Default seed for MRG31K3P PRNG.
Definition: rocrand_mrg31k3p.h:46
__forceinline__ __device__ __host__ unsigned int rocrand(rocrand_state_mrg31k3p *state)
Returns uniformly distributed random unsigned int value from [0; 2^32 - 1] range.
Definition: rocrand_mrg31k3p.h:368
__forceinline__ __device__ __host__ void skipahead_subsequence(unsigned long long subsequence, rocrand_state_mrg31k3p *state)
Updates MRG31K3P state to skip ahead by subsequence subsequences.
Definition: rocrand_mrg31k3p.h:398
__forceinline__ __device__ __host__ void skipahead(unsigned long long offset, rocrand_state_mrg31k3p *state)
Updates MRG31K3P state to skip ahead by offset elements.
Definition: rocrand_mrg31k3p.h:383
__forceinline__ __device__ __host__ void skipahead_sequence(unsigned long long sequence, rocrand_state_mrg31k3p *state)
Updates MRG31K3P state to skip ahead by sequence sequences.
Definition: rocrand_mrg31k3p.h:413
__forceinline__ __device__ __host__ void rocrand_init(const unsigned long long seed, const unsigned long long subsequence, const unsigned long long offset, rocrand_state_mrg31k3p *state)
Initializes MRG31K3P state.
Definition: rocrand_mrg31k3p.h:347