/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-rocrand/checkouts/docs-7.1.0/projects/rocrand/library/include/rocrand/rocrand_mrg32k3a.h Source File

/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-rocrand/checkouts/docs-7.1.0/projects/rocrand/library/include/rocrand/rocrand_mrg32k3a.h Source File#

API library: /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-rocrand/checkouts/docs-7.1.0/projects/rocrand/library/include/rocrand/rocrand_mrg32k3a.h Source File
 // Copyright (c) 2017-2025 Advanced Micro Devices, Inc. All rights reserved.
 //
 // Permission is hereby granted, free of charge, to any person obtaining a copy
 // of this software and associated documentation files (the "Software"), to deal
 // in the Software without restriction, including without limitation the rights
 // to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 // copies of the Software, and to permit persons to whom the Software is
 // furnished to do so, subject to the following conditions:
 //
 // The above copyright notice and this permission notice shall be included in
 // all copies or substantial portions of the Software.
 //
 // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 // IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 // FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
 // AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 // LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 // THE SOFTWARE.
  
 #ifndef ROCRAND_MRG32K3A_H_
 #define ROCRAND_MRG32K3A_H_
  
 #include "rocrand/rocrand_common.h"
 #include "rocrand/rocrand_mrg32k3a_precomputed.h"
  
 #include <hip/hip_runtime.h>
  
 #define ROCRAND_MRG32K3A_POW32 4294967296U
 #define ROCRAND_MRG32K3A_M1 4294967087U
 #define ROCRAND_MRG32K3A_M1C 209U
 #define ROCRAND_MRG32K3A_M2 4294944443U
 #define ROCRAND_MRG32K3A_M2C 22853U
 #define ROCRAND_MRG32K3A_A12 1403580U
 #define ROCRAND_MRG32K3A_A13 (4294967087U - 810728U)
 #define ROCRAND_MRG32K3A_A13N 810728U
 #define ROCRAND_MRG32K3A_A21 527612U
 #define ROCRAND_MRG32K3A_A23 (4294944443U - 1370589U)
 #define ROCRAND_MRG32K3A_A23N 1370589U
 #define ROCRAND_MRG32K3A_NORM_DOUBLE (2.3283065498378288e-10) // 1/ROCRAND_MRG32K3A_M1
 #define ROCRAND_MRG32K3A_UINT_NORM \
     (1.000000048661607) // (ROCRAND_MRG32K3A_POW32 - 1)/(ROCRAND_MRG32K3A_M1 - 1)
  
  #define ROCRAND_MRG32K3A_DEFAULT_SEED 12345ULL // end of group rocranddevice
  
 namespace rocrand_device {
  
 class mrg32k3a_engine
 {
 public:
     struct mrg32k3a_state
     {
         unsigned int g1[3];
         unsigned int g2[3];
  
     #ifndef ROCRAND_DETAIL_BM_NOT_IN_STATE
         // The Box–Muller transform requires two inputs to convert uniformly
         // distributed real values [0; 1] to normally distributed real values
         // (with mean = 0, and stddev = 1). Often user wants only one
         // normally distributed number, to save performance and random
         // numbers the 2nd value is saved for future requests.
         unsigned int boxmuller_float_state; // is there a float in boxmuller_float
         unsigned int boxmuller_double_state; // is there a double in boxmuller_double
         float boxmuller_float; // normally distributed float
         double boxmuller_double; // normally distributed double
     #endif
     };
  
     __forceinline__ __device__ __host__ mrg32k3a_engine()
     {
         this->seed(ROCRAND_MRG32K3A_DEFAULT_SEED, 0, 0);
     }
  
     __forceinline__ __device__ __host__ mrg32k3a_engine(const unsigned long long seed,
                                                         const unsigned long long subsequence,
                                                         const unsigned long long offset)
     {
         this->seed(seed, subsequence, offset);
     }
  
     __forceinline__ __device__ __host__ void seed(unsigned long long       seed_value,
                                                   const unsigned long long subsequence,
                                                   const unsigned long long offset)
     {
         if(seed_value == 0)
         {
             seed_value = ROCRAND_MRG32K3A_DEFAULT_SEED;
         }
         unsigned int x = (unsigned int) seed_value ^ 0x55555555U;
         unsigned int y = (unsigned int) ((seed_value >> 32) ^ 0xAAAAAAAAU);
         m_state.g1[0] = mod_mul_m1(x, seed_value);
         m_state.g1[1] = mod_mul_m1(y, seed_value);
         m_state.g1[2] = mod_mul_m1(x, seed_value);
         m_state.g2[0] = mod_mul_m2(y, seed_value);
         m_state.g2[1] = mod_mul_m2(x, seed_value);
         m_state.g2[2] = mod_mul_m2(y, seed_value);
         this->restart(subsequence, offset);
     }
  
     __forceinline__ __device__ __host__ void discard(unsigned long long offset)
     {
         this->discard_impl(offset);
     }
  
     __forceinline__ __device__ __host__ void discard_subsequence(unsigned long long subsequence)
     {
         this->discard_subsequence_impl(subsequence);
     }
  
     __forceinline__ __device__ __host__ void discard_sequence(unsigned long long sequence)
     {
         this->discard_sequence_impl(sequence);
     }
  
     __forceinline__ __device__ __host__ void restart(const unsigned long long subsequence,
                                                      const unsigned long long offset)
     {
     #ifndef ROCRAND_DETAIL_BM_NOT_IN_STATE
         m_state.boxmuller_float_state = 0;
         m_state.boxmuller_double_state = 0;
     #endif
         this->discard_subsequence_impl(subsequence);
         this->discard_impl(offset);
     }
  
     __forceinline__ __device__ __host__ unsigned int operator()()
     {
         return this->next();
     }
  
     // Returned value is in range [1, ROCRAND_MRG32K3A_M1],
     // where ROCRAND_MRG32K3A_M1 < UINT_MAX
     __forceinline__ __device__ __host__
     unsigned int next()
     {
         const unsigned int p1 = mod_m1(detail::mad_u64_u32(
             ROCRAND_MRG32K3A_A12,
             m_state.g1[1],
             detail::mul_u64_u32(ROCRAND_MRG32K3A_A13N, (ROCRAND_MRG32K3A_M1 - m_state.g1[0]))));
  
         m_state.g1[0] = m_state.g1[1];
         m_state.g1[1] = m_state.g1[2];
         m_state.g1[2] = p1;
  
         const unsigned int p2 = mod_m2(detail::mad_u64_u32(
             ROCRAND_MRG32K3A_A21,
             m_state.g2[2],
             detail::mul_u64_u32(ROCRAND_MRG32K3A_A23N, (ROCRAND_MRG32K3A_M2 - m_state.g2[0]))));
  
         m_state.g2[0] = m_state.g2[1];
         m_state.g2[1] = m_state.g2[2];
         m_state.g2[2] = p2;
  
         return (p1 - p2) + (p1 <= p2 ? ROCRAND_MRG32K3A_M1 : 0);
     }
  
 protected:
     // Advances the internal state to skip \p offset numbers.
     // DOES NOT CALCULATE NEW ULONGLONG
     __forceinline__ __device__ __host__ void discard_impl(unsigned long long offset)
     {
         discard_state(offset);
     }
  
     // DOES NOT CALCULATE NEW ULONGLONG
     __forceinline__ __device__ __host__ void
         discard_subsequence_impl(unsigned long long subsequence)
     {
         int i = 0;
  
         while(subsequence > 0) {
             if (subsequence & 1) {
                 #if defined(__HIP_DEVICE_COMPILE__)
                 mod_mat_vec_m1(d_A1P76 + i, m_state.g1);
                 mod_mat_vec_m2(d_A2P76 + i, m_state.g2);
                 #else
                 mod_mat_vec_m1(h_A1P76 + i, m_state.g1);
                 mod_mat_vec_m2(h_A2P76 + i, m_state.g2);
                 #endif
             }
             subsequence >>= 1;
             i += 9;
         }
     }
  
     // DOES NOT CALCULATE NEW ULONGLONG
     __forceinline__ __device__ __host__ void discard_sequence_impl(unsigned long long sequence)
     {
         int i = 0;
  
         while(sequence > 0) {
             if (sequence & 1) {
                 #if defined(__HIP_DEVICE_COMPILE__)
                 mod_mat_vec_m1(d_A1P127 + i, m_state.g1);
                 mod_mat_vec_m2(d_A2P127 + i, m_state.g2);
                 #else
                 mod_mat_vec_m1(h_A1P127 + i, m_state.g1);
                 mod_mat_vec_m2(h_A2P127 + i, m_state.g2);
                 #endif
             }
             sequence >>= 1;
             i += 9;
         }
     }
  
     // Advances the internal state by offset times.
     // DOES NOT CALCULATE NEW ULONGLONG
     __forceinline__ __device__ __host__ void discard_state(unsigned long long offset)
     {
         int i = 0;
  
         while(offset > 0) {
             if (offset & 1) {
                 #if defined(__HIP_DEVICE_COMPILE__)
                 mod_mat_vec_m1(d_A1 + i, m_state.g1);
                 mod_mat_vec_m2(d_A2 + i, m_state.g2);
                 #else
                 mod_mat_vec_m1(h_A1 + i, m_state.g1);
                 mod_mat_vec_m2(h_A2 + i, m_state.g2);
                 #endif
             }
             offset >>= 1;
             i += 9;
         }
     }
  
     // Advances the internal state to the next state
     // DOES NOT CALCULATE NEW ULONGLONG
     __forceinline__ __device__ __host__ void discard_state()
     {
         discard_state(1);
     }
  
 private:
     __forceinline__ __device__ __host__
     static void mod_mat_vec_m1(const unsigned int* A, unsigned int* s)
     {
         unsigned long long x[3] = {s[0], s[1], s[2]};
  
         s[0] = mod_m1(mod_m1(A[0] * x[0]) + mod_m1(A[1] * x[1]) + mod_m1(A[2] * x[2]));
  
         s[1] = mod_m1(mod_m1(A[3] * x[0]) + mod_m1(A[4] * x[1]) + mod_m1(A[5] * x[2]));
  
         s[2] = mod_m1(mod_m1(A[6] * x[0]) + mod_m1(A[7] * x[1]) + mod_m1(A[8] * x[2]));
     }
  
     __forceinline__ __device__ __host__
     static void mod_mat_vec_m2(const unsigned int* A, unsigned int* s)
     {
         unsigned long long x[3] = {s[0], s[1], s[2]};
  
         s[0] = mod_m2(mod_m2(A[0] * x[0]) + mod_m2(A[1] * x[1]) + mod_m2(A[2] * x[2]));
  
         s[1] = mod_m2(mod_m2(A[3] * x[0]) + mod_m2(A[4] * x[1]) + mod_m2(A[5] * x[2]));
  
         s[2] = mod_m2(mod_m2(A[6] * x[0]) + mod_m2(A[7] * x[1]) + mod_m2(A[8] * x[2]));
     }
  
     __forceinline__ __device__ __host__ static unsigned long long mod_mul_m1(unsigned int       i,
                                                                              unsigned long long j)
     {
         long long hi, lo, temp1, temp2;
  
         hi = i / 131072;
         lo = i - (hi * 131072);
         temp1 = mod_m1(hi * j) * 131072;
         temp2 = mod_m1(lo * j);
         lo = mod_m1(temp1 + temp2);
  
         if (lo < 0)
             lo += ROCRAND_MRG32K3A_M1;
         return lo;
     }
  
     __forceinline__ __device__ __host__
     static unsigned long long mod_m1(unsigned long long p)
     {
         p = detail::mad_u64_u32(ROCRAND_MRG32K3A_M1C,
                                 static_cast<unsigned int>(p >> 32),
                                 static_cast<unsigned int>(p));
         if(p >= ROCRAND_MRG32K3A_M1)
             p -= ROCRAND_MRG32K3A_M1;
  
         return p;
     }
  
     __forceinline__ __device__ __host__
     static unsigned long long mod_mul_m2(unsigned int i, unsigned long long j)
     {
         long long hi, lo, temp1, temp2;
  
         hi = i / 131072;
         lo = i - (hi * 131072);
         temp1 = mod_m2(hi * j) * 131072;
         temp2 = mod_m2(lo * j);
         lo = mod_m2(temp1 + temp2);
  
         if (lo < 0)
             lo += ROCRAND_MRG32K3A_M2;
         return lo;
     }
  
     __forceinline__ __device__ __host__
     static unsigned long long mod_m2(unsigned long long p)
     {
         p = detail::mad_u64_u32(ROCRAND_MRG32K3A_M2C,
                                 static_cast<unsigned int>(p >> 32),
                                 static_cast<unsigned int>(p));
         p = detail::mad_u64_u32(ROCRAND_MRG32K3A_M2C,
                                 static_cast<unsigned int>(p >> 32),
                                 static_cast<unsigned int>(p));
         if(p >= ROCRAND_MRG32K3A_M2)
             p -= ROCRAND_MRG32K3A_M2;
  
         return p;
     }
  
 protected:
     // State
     mrg32k3a_state m_state;
  
     #ifndef ROCRAND_DETAIL_BM_NOT_IN_STATE
     friend struct detail::engine_boxmuller_helper<mrg32k3a_engine>;
     #endif
  
 }; // mrg32k3a_engine class
  
 } // end namespace rocrand_device
  
 typedef rocrand_device::mrg32k3a_engine rocrand_state_mrg32k3a;
  
 __forceinline__ __device__ __host__
 void rocrand_init(const unsigned long long seed,
                   const unsigned long long subsequence,
                   const unsigned long long offset,
                   rocrand_state_mrg32k3a*  state)
 {
     *state = rocrand_state_mrg32k3a(seed, subsequence, offset);
 }
  
 __forceinline__ __device__ __host__
 unsigned int rocrand(rocrand_state_mrg32k3a* state)
 {
     // next() in [1, ROCRAND_MRG32K3A_M1]
     return static_cast<unsigned int>((state->next() - 1) * ROCRAND_MRG32K3A_UINT_NORM);
 }
  
 __forceinline__ __device__ __host__
 void skipahead(unsigned long long offset, rocrand_state_mrg32k3a* state)
 {
     return state->discard(offset);
 }
  
 __forceinline__ __device__ __host__
 void skipahead_subsequence(unsigned long long subsequence, rocrand_state_mrg32k3a* state)
 {
     return state->discard_subsequence(subsequence);
 }
  
 __forceinline__ __device__ __host__
 void skipahead_sequence(unsigned long long sequence, rocrand_state_mrg32k3a* state)
 {
     return state->discard_sequence(sequence);
 }
  // end of group rocranddevice
  
 #endif // ROCRAND_MRG32K3A_H_