1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
|
/***************************************************************************
* Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and *
* Martin Renou *
* Copyright (c) QuantStack *
* Copyright (c) Serge Guelton *
* *
* Distributed under the terms of the BSD 3-Clause License. *
* *
* The full license is in the file LICENSE, distributed with this software. *
****************************************************************************/
#ifndef XSIMD_SSE4_1_HPP
#define XSIMD_SSE4_1_HPP
#include <type_traits>
#include "../types/xsimd_sse4_1_register.hpp"
namespace xsimd
{
namespace kernel
{
using namespace types;
// any
template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
inline bool any(batch<T, A> const& self, requires_arch<sse4_1>) noexcept
{
return !_mm_testz_si128(self, self);
}
// ceil
template <class A>
inline batch<float, A> ceil(batch<float, A> const& self, requires_arch<sse4_1>) noexcept
{
return _mm_ceil_ps(self);
}
template <class A>
inline batch<double, A> ceil(batch<double, A> const& self, requires_arch<sse4_1>) noexcept
{
return _mm_ceil_pd(self);
}
// fast_cast
namespace detail
{
template <class A>
inline batch<double, A> fast_cast(batch<int64_t, A> const& x, batch<double, A> const&, requires_arch<sse4_1>) noexcept
{
// from https://stackoverflow.com/questions/41144668/how-to-efficiently-perform-double-int64-conversions-with-sse-avx
__m128i xH = _mm_srai_epi32(x, 16);
xH = _mm_blend_epi16(xH, _mm_setzero_si128(), 0x33);
xH = _mm_add_epi64(xH, _mm_castpd_si128(_mm_set1_pd(442721857769029238784.))); // 3*2^67
__m128i xL = _mm_blend_epi16(x, _mm_castpd_si128(_mm_set1_pd(0x0010000000000000)), 0x88); // 2^52
__m128d f = _mm_sub_pd(_mm_castsi128_pd(xH), _mm_set1_pd(442726361368656609280.)); // 3*2^67 + 2^52
return _mm_add_pd(f, _mm_castsi128_pd(xL));
}
template <class A>
inline batch<double, A> fast_cast(batch<uint64_t, A> const& x, batch<double, A> const&, requires_arch<sse4_1>) noexcept
{
// from https://stackoverflow.com/questions/41144668/how-to-efficiently-perform-double-int64-conversions-with-sse-avx
__m128i xH = _mm_srli_epi64(x, 32);
xH = _mm_or_si128(xH, _mm_castpd_si128(_mm_set1_pd(19342813113834066795298816.))); // 2^84
__m128i xL = _mm_blend_epi16(x, _mm_castpd_si128(_mm_set1_pd(0x0010000000000000)), 0xcc); // 2^52
__m128d f = _mm_sub_pd(_mm_castsi128_pd(xH), _mm_set1_pd(19342813118337666422669312.)); // 2^84 + 2^52
return _mm_add_pd(f, _mm_castsi128_pd(xL));
}
}
// eq
template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
inline batch_bool<T, A> eq(batch<T, A> const& self, batch<T, A> const& other, requires_arch<sse4_1>) noexcept
{
XSIMD_IF_CONSTEXPR(sizeof(T) == 8)
{
return _mm_cmpeq_epi64(self, other);
}
else
{
return eq(self, other, ssse3 {});
}
}
// floor
template <class A>
inline batch<float, A> floor(batch<float, A> const& self, requires_arch<sse4_1>) noexcept
{
return _mm_floor_ps(self);
}
template <class A>
inline batch<double, A> floor(batch<double, A> const& self, requires_arch<sse4_1>) noexcept
{
return _mm_floor_pd(self);
}
// insert
template <class A, class T, size_t I, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
inline batch<T, A> insert(batch<T, A> const& self, T val, index<I> pos, requires_arch<sse4_1>) noexcept
{
XSIMD_IF_CONSTEXPR(sizeof(T) == 1)
{
return _mm_insert_epi8(self, val, I);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
{
return _mm_insert_epi32(self, val, I);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)
{
#if (!defined(_MSC_VER) && __x86_64__) || (_MSC_VER > 1900 && defined(_M_X64))
return _mm_insert_epi64(self, val, I);
#else
uint32_t lo, hi;
memcpy(&lo, (reinterpret_cast<uint32_t*>(&val)), sizeof(lo));
memcpy(&hi, (reinterpret_cast<uint32_t*>(&val)) + 1, sizeof(hi));
return _mm_insert_epi32(_mm_insert_epi32(self, lo, 2 * I), hi, 2 * I + 1);
#endif
}
else
{
return insert(self, val, pos, ssse3 {});
}
}
// max
template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
inline batch<T, A> max(batch<T, A> const& self, batch<T, A> const& other, requires_arch<sse4_1>) noexcept
{
if (std::is_signed<T>::value)
{
XSIMD_IF_CONSTEXPR(sizeof(T) == 1)
{
return _mm_max_epi8(self, other);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
{
return _mm_max_epi16(self, other);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
{
return _mm_max_epi32(self, other);
}
else
{
return max(self, other, ssse3 {});
}
}
else
{
XSIMD_IF_CONSTEXPR(sizeof(T) == 1)
{
return _mm_max_epu8(self, other);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
{
return _mm_max_epu16(self, other);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
{
return _mm_max_epu32(self, other);
}
else
{
return max(self, other, ssse3 {});
}
}
}
// min
template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
inline batch<T, A> min(batch<T, A> const& self, batch<T, A> const& other, requires_arch<sse4_1>) noexcept
{
if (std::is_signed<T>::value)
{
XSIMD_IF_CONSTEXPR(sizeof(T) == 1)
{
return _mm_min_epi8(self, other);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
{
return _mm_min_epi16(self, other);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
{
return _mm_min_epi32(self, other);
}
else
{
return min(self, other, ssse3 {});
}
}
else
{
XSIMD_IF_CONSTEXPR(sizeof(T) == 1)
{
return _mm_min_epu8(self, other);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
{
return _mm_min_epu16(self, other);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
{
return _mm_min_epu32(self, other);
}
else
{
return min(self, other, ssse3 {});
}
}
}
// mul
template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
inline batch<T, A> mul(batch<T, A> const& self, batch<T, A> const& other, requires_arch<sse4_1>) noexcept
{
XSIMD_IF_CONSTEXPR(sizeof(T) == 1)
{
return _mm_or_si128(
_mm_and_si128(_mm_mullo_epi16(self, other), _mm_srli_epi16(_mm_cmpeq_epi8(self, self), 8)),
_mm_slli_epi16(_mm_mullo_epi16(_mm_srli_epi16(self, 8), _mm_srli_epi16(other, 8)), 8));
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
{
return _mm_mullo_epi16(self, other);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
{
return _mm_mullo_epi32(self, other);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)
{
return _mm_add_epi64(
_mm_mul_epu32(self, other),
_mm_slli_epi64(
_mm_add_epi64(
_mm_mul_epu32(other, _mm_shuffle_epi32(self, _MM_SHUFFLE(2, 3, 0, 1))),
_mm_mul_epu32(self, _mm_shuffle_epi32(other, _MM_SHUFFLE(2, 3, 0, 1)))),
32));
}
else
{
assert(false && "unsupported arch/op combination");
return {};
}
}
// nearbyint
template <class A>
inline batch<float, A> nearbyint(batch<float, A> const& self, requires_arch<sse4_1>) noexcept
{
return _mm_round_ps(self, _MM_FROUND_TO_NEAREST_INT);
}
template <class A>
inline batch<double, A> nearbyint(batch<double, A> const& self, requires_arch<sse4_1>) noexcept
{
return _mm_round_pd(self, _MM_FROUND_TO_NEAREST_INT);
}
// select
namespace detail
{
template <class T>
inline constexpr T interleave(T const& cond) noexcept
{
return (((cond * 0x0101010101010101ULL & 0x8040201008040201ULL) * 0x0102040810204081ULL >> 49) & 0x5555) | (((cond * 0x0101010101010101ULL & 0x8040201008040201ULL) * 0x0102040810204081ULL >> 48) & 0xAAAA);
}
}
template <class A, class T, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
inline batch<T, A> select(batch_bool<T, A> const& cond, batch<T, A> const& true_br, batch<T, A> const& false_br, requires_arch<sse4_1>) noexcept
{
return _mm_blendv_epi8(false_br, true_br, cond);
}
template <class A>
inline batch<float, A> select(batch_bool<float, A> const& cond, batch<float, A> const& true_br, batch<float, A> const& false_br, requires_arch<sse4_1>) noexcept
{
return _mm_blendv_ps(false_br, true_br, cond);
}
template <class A>
inline batch<double, A> select(batch_bool<double, A> const& cond, batch<double, A> const& true_br, batch<double, A> const& false_br, requires_arch<sse4_1>) noexcept
{
return _mm_blendv_pd(false_br, true_br, cond);
}
template <class A, class T, bool... Values, class = typename std::enable_if<std::is_integral<T>::value, void>::type>
inline batch<T, A> select(batch_bool_constant<batch<T, A>, Values...> const&, batch<T, A> const& true_br, batch<T, A> const& false_br, requires_arch<sse4_1>) noexcept
{
constexpr int mask = batch_bool_constant<batch<T, A>, Values...>::mask();
XSIMD_IF_CONSTEXPR(sizeof(T) == 2)
{
return _mm_blend_epi16(false_br, true_br, mask);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 4)
{
constexpr int imask = detail::interleave(mask);
return _mm_blend_epi16(false_br, true_br, imask);
}
else XSIMD_IF_CONSTEXPR(sizeof(T) == 8)
{
constexpr int imask = detail::interleave(mask);
constexpr int imask2 = detail::interleave(imask);
return _mm_blend_epi16(false_br, true_br, imask2);
}
else
{
return select(batch_bool_constant<batch<T, A>, Values...>(), true_br, false_br, ssse3 {});
}
}
template <class A, bool... Values>
inline batch<float, A> select(batch_bool_constant<batch<float, A>, Values...> const&, batch<float, A> const& true_br, batch<float, A> const& false_br, requires_arch<sse4_1>) noexcept
{
constexpr int mask = batch_bool_constant<batch<float, A>, Values...>::mask();
return _mm_blend_ps(false_br, true_br, mask);
}
template <class A, bool... Values>
inline batch<double, A> select(batch_bool_constant<batch<double, A>, Values...> const&, batch<double, A> const& true_br, batch<double, A> const& false_br, requires_arch<sse4_1>) noexcept
{
constexpr int mask = batch_bool_constant<batch<double, A>, Values...>::mask();
return _mm_blend_pd(false_br, true_br, mask);
}
// trunc
template <class A>
inline batch<float, A> trunc(batch<float, A> const& self, requires_arch<sse4_1>) noexcept
{
return _mm_round_ps(self, _MM_FROUND_TO_ZERO);
}
template <class A>
inline batch<double, A> trunc(batch<double, A> const& self, requires_arch<sse4_1>) noexcept
{
return _mm_round_pd(self, _MM_FROUND_TO_ZERO);
}
}
}
#endif
|
