1// Optimizations for random number extensions, x86 version -*- C++ -*-
2
3// Copyright (C) 2012-2020 Free Software Foundation, Inc.
4//
5// This file is part of the GNU ISO C++ Library.  This library is free
6// software; you can redistribute it and/or modify it under the
7// terms of the GNU General Public License as published by the
8// Free Software Foundation; either version 3, or (at your option)
9// any later version.
10
11// This library is distributed in the hope that it will be useful,
12// but WITHOUT ANY WARRANTY; without even the implied warranty of
13// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14// GNU General Public License for more details.
15
16// Under Section 7 of GPL version 3, you are granted additional
17// permissions described in the GCC Runtime Library Exception, version
18// 3.1, as published by the Free Software Foundation.
19
20// You should have received a copy of the GNU General Public License and
21// a copy of the GCC Runtime Library Exception along with this program;
22// see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
23// <http://www.gnu.org/licenses/>.
24
25/** @file ext/random.tcc
26 *  This is an internal header file, included by other library headers.
27 *  Do not attempt to use it directly. @headername{ext/random}
28 */
29
30#ifndef _EXT_OPT_RANDOM_H
31#define _EXT_OPT_RANDOM_H 1
32
33#pragma GCC system_header
34
35#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
36
37#ifdef __SSE2__
38
39namespace __gnu_cxx _GLIBCXX_VISIBILITY(default)
40{
41_GLIBCXX_BEGIN_NAMESPACE_VERSION
42
43  namespace {
44
45    template<size_t __sl1, size_t __sl2, size_t __sr1, size_t __sr2,
46	     uint32_t __msk1, uint32_t __msk2, uint32_t __msk3, uint32_t __msk4>
47      inline __m128i __sse2_recursion(__m128i __a, __m128i __b,
48				      __m128i __c, __m128i __d)
49      {
50	__m128i __y = _mm_srli_epi32(__b, __sr1);
51	__m128i __z = _mm_srli_si128(__c, __sr2);
52	__m128i __v = _mm_slli_epi32(__d, __sl1);
53	__z = _mm_xor_si128(__z, __a);
54	__z = _mm_xor_si128(__z, __v);
55	__m128i __x = _mm_slli_si128(__a, __sl2);
56	__y = _mm_and_si128(__y, _mm_set_epi32(__msk4, __msk3, __msk2, __msk1));
57	__z = _mm_xor_si128(__z, __x);
58	return _mm_xor_si128(__z, __y);
59      }
60
61  }
62
63
64#define _GLIBCXX_OPT_HAVE_RANDOM_SFMT_GEN_READ	1
65  template<typename _UIntType, size_t __m,
66	   size_t __pos1, size_t __sl1, size_t __sl2,
67	   size_t __sr1, size_t __sr2,
68	   uint32_t __msk1, uint32_t __msk2,
69	   uint32_t __msk3, uint32_t __msk4,
70	   uint32_t __parity1, uint32_t __parity2,
71	   uint32_t __parity3, uint32_t __parity4>
72    void simd_fast_mersenne_twister_engine<_UIntType, __m,
73					   __pos1, __sl1, __sl2, __sr1, __sr2,
74					   __msk1, __msk2, __msk3, __msk4,
75					   __parity1, __parity2, __parity3,
76					   __parity4>::
77    _M_gen_rand(void)
78    {
79      __m128i __r1 = _mm_load_si128(&_M_state[_M_nstate - 2]);
80      __m128i __r2 = _mm_load_si128(&_M_state[_M_nstate - 1]);
81
82      size_t __i;
83      for (__i = 0; __i < _M_nstate - __pos1; ++__i)
84	{
85	  __m128i __r = __sse2_recursion<__sl1, __sl2, __sr1, __sr2,
86					 __msk1, __msk2, __msk3, __msk4>
87	    (_M_state[__i], _M_state[__i + __pos1], __r1, __r2);
88	  _mm_store_si128(&_M_state[__i], __r);
89	  __r1 = __r2;
90	  __r2 = __r;
91	}
92      for (; __i < _M_nstate; ++__i)
93	{
94	  __m128i __r = __sse2_recursion<__sl1, __sl2, __sr1, __sr2,
95					 __msk1, __msk2, __msk3, __msk4>
96	    (_M_state[__i], _M_state[__i + __pos1 - _M_nstate], __r1, __r2);
97	  _mm_store_si128(&_M_state[__i], __r);
98	  __r1 = __r2;
99	  __r2 = __r;
100	}
101
102      _M_pos = 0;
103    }
104
105
106#define _GLIBCXX_OPT_HAVE_RANDOM_SFMT_OPERATOREQUAL	1
107  template<typename _UIntType, size_t __m,
108	   size_t __pos1, size_t __sl1, size_t __sl2,
109	   size_t __sr1, size_t __sr2,
110	   uint32_t __msk1, uint32_t __msk2,
111	   uint32_t __msk3, uint32_t __msk4,
112	   uint32_t __parity1, uint32_t __parity2,
113	   uint32_t __parity3, uint32_t __parity4>
114    bool
115    operator==(const __gnu_cxx::simd_fast_mersenne_twister_engine<_UIntType,
116	       __m, __pos1, __sl1, __sl2, __sr1, __sr2,
117	       __msk1, __msk2, __msk3, __msk4,
118	       __parity1, __parity2, __parity3, __parity4>& __lhs,
119	       const __gnu_cxx::simd_fast_mersenne_twister_engine<_UIntType,
120	       __m, __pos1, __sl1, __sl2, __sr1, __sr2,
121	       __msk1, __msk2, __msk3, __msk4,
122	       __parity1, __parity2, __parity3, __parity4>& __rhs)
123    {
124      __m128i __res = _mm_cmpeq_epi8(__lhs._M_state[0], __rhs._M_state[0]);
125      for (size_t __i = 1; __i < __lhs._M_nstate; ++__i)
126	__res = _mm_and_si128(__res, _mm_cmpeq_epi8(__lhs._M_state[__i],
127						    __rhs._M_state[__i]));
128      return (_mm_movemask_epi8(__res) == 0xffff
129	      && __lhs._M_pos == __rhs._M_pos);
130    }
131
132
133_GLIBCXX_END_NAMESPACE_VERSION
134} // namespace
135
136#endif // __SSE2__
137
138#endif // __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
139
140#endif // _EXT_OPT_RANDOM_H
141