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
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
|
/*
* Simd Library (http://ermig1979.github.io/Simd).
*
* Copyright (c) 2011-2022 Yermalayeu Ihar.
*
* 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 __SimdStore_h__
#define __SimdStore_h__
#include "Simd/SimdDefs.h"
#include "Simd/SimdLoad.h"
#include "Simd/SimdMath.h"
#include "Simd/SimdLog.h"
namespace Simd
{
#ifdef SIMD_SSE41_ENABLE
namespace Sse41
{
template <bool align> SIMD_INLINE void Store(float* p, __m128 a);
template <> SIMD_INLINE void Store<false>(float* p, __m128 a)
{
_mm_storeu_ps(p, a);
}
template <> SIMD_INLINE void Store<true>(float* p, __m128 a)
{
_mm_store_ps(p, a);
}
template <int part> SIMD_INLINE void StoreHalf(float* p, __m128 a);
template <> SIMD_INLINE void StoreHalf<0>(float* p, __m128 a)
{
_mm_storel_pi((__m64*)p, a);
}
template <> SIMD_INLINE void StoreHalf<1>(float* p, __m128 a)
{
_mm_storeh_pi((__m64*)p, a);
}
SIMD_INLINE void Store(float* ptr, __m128 val, size_t size)
{
SIMD_ALIGNED(16) float buf[F];
_mm_store_ps(buf, val);
for (size_t i = 0; i < size; ++i)
ptr[i] = buf[i];
}
template<int step> SIMD_INLINE void Scater(float* ptr, __m128 val)
{
SIMD_ALIGNED(16) float buf[F];
_mm_store_ps(buf, val);
ptr[0 * step] = buf[0];
ptr[1 * step] = buf[1];
ptr[2 * step] = buf[2];
ptr[3 * step] = buf[3];
}
template<int step> SIMD_INLINE void Scater(float* ptr, __m128 val, size_t size)
{
SIMD_ALIGNED(16) float buf[F];
_mm_store_ps(buf, val);
for (size_t i = 0; i < size; ++i)
ptr[i * step] = buf[i];
}
template <bool align> SIMD_INLINE void Store(__m128i * p, __m128i a);
template <> SIMD_INLINE void Store<false>(__m128i * p, __m128i a)
{
_mm_storeu_si128(p, a);
}
template <> SIMD_INLINE void Store<true>(__m128i * p, __m128i a)
{
_mm_store_si128(p, a);
}
template <int part> SIMD_INLINE void StoreHalf(__m128i * p, __m128i a)
{
StoreHalf<part>((float*)p, _mm_castsi128_ps(a));
}
template <bool align> SIMD_INLINE void StoreMasked(float* p, __m128 value, __m128 mask)
{
__m128 old = Load<align>(p);
Store<align>(p, _mm_blendv_ps(old, value, mask));
}
template <bool align> SIMD_INLINE void StoreMasked(__m128i* p, __m128i value, __m128i mask)
{
__m128i old = Load<align>(p);
Store<align>(p, _mm_blendv_epi8(old, value, mask));
}
}
#endif
#ifdef SIMD_AVX_ENABLE
namespace Avx
{
template <bool align> SIMD_INLINE void Store(float * p, __m256 a);
template <> SIMD_INLINE void Store<false>(float * p, __m256 a)
{
_mm256_storeu_ps(p, a);
}
template <> SIMD_INLINE void Store<true>(float * p, __m256 a)
{
_mm256_store_ps(p, a);
}
SIMD_INLINE void Store(float* ptr, __m256 val, size_t size)
{
SIMD_ALIGNED(32) float buf[F];
_mm256_store_ps(buf, val);
for (size_t i = 0; i < size; ++i)
ptr[i] = buf[i];
}
template <bool align> SIMD_INLINE void Store(float * p0, float * p1, __m256 a)
{
Sse41::Store<align>(p0, _mm256_extractf128_ps(a, 0));
Sse41::Store<align>(p1, _mm256_extractf128_ps(a, 1));
}
template <bool align> SIMD_INLINE void StoreMasked(float * p, __m256 value, __m256 mask)
{
__m256 old = Load<align>(p);
Store<align>(p, _mm256_blendv_ps(old, value, mask));
}
template<int step> SIMD_INLINE void Scater(float* ptr, __m256 val)
{
SIMD_ALIGNED(32) float buf[F];
_mm256_store_ps(buf, val);
ptr[0 * step] = buf[0];
ptr[1 * step] = buf[1];
ptr[2 * step] = buf[2];
ptr[3 * step] = buf[3];
ptr[4 * step] = buf[4];
ptr[5 * step] = buf[5];
ptr[6 * step] = buf[6];
ptr[7 * step] = buf[7];
}
template<int step> SIMD_INLINE void Scater(float* ptr, __m256 val, size_t size)
{
SIMD_ALIGNED(32) float buf[F];
_mm256_store_ps(buf, val);
for (size_t i = 0; i < size; ++i)
ptr[i * step] = buf[i];
}
}
#endif
#ifdef SIMD_AVX2_ENABLE
namespace Avx2
{
using namespace Avx;
template <bool align> SIMD_INLINE void Store(__m256i * p, __m256i a);
template <> SIMD_INLINE void Store<false>(__m256i * p, __m256i a)
{
_mm256_storeu_si256(p, a);
}
template <> SIMD_INLINE void Store<true>(__m256i * p, __m256i a)
{
_mm256_store_si256(p, a);
}
template <bool align> SIMD_INLINE void Store(__m128i* p0, __m128i* p1, __m256i a)
{
Sse41::Store<align>(p0, _mm256_extractf128_si256(a, 0));
Sse41::Store<align>(p1, _mm256_extractf128_si256(a, 1));
}
template <bool align> SIMD_INLINE void StoreMasked(__m256i * p, __m256i value, __m256i mask)
{
__m256i old = Load<align>(p);
Store<align>(p, _mm256_blendv_epi8(old, value, mask));
}
SIMD_INLINE __m256i PackI16ToI8(__m256i lo, __m256i hi)
{
return _mm256_permute4x64_epi64(_mm256_packs_epi16(lo, hi), 0xD8);
}
SIMD_INLINE __m256i PackI16ToU8(__m256i lo, __m256i hi)
{
return _mm256_permute4x64_epi64(_mm256_packus_epi16(lo, hi), 0xD8);
}
SIMD_INLINE __m256i PackI32ToI16(__m256i lo, __m256i hi)
{
return _mm256_permute4x64_epi64(_mm256_packs_epi32(lo, hi), 0xD8);
}
SIMD_INLINE __m256i PackU32ToI16(__m256i lo, __m256i hi)
{
return _mm256_permute4x64_epi64(_mm256_packus_epi32(lo, hi), 0xD8);
}
SIMD_INLINE void Permute2x128(__m256i & lo, __m256i & hi)
{
__m256i _lo = lo;
lo = _mm256_permute2x128_si256(lo, hi, 0x20);
hi = _mm256_permute2x128_si256(_lo, hi, 0x31);
}
template <bool align> SIMD_INLINE void Store24(uint8_t * p, __m256i a)
{
Sse41::Store<align>((__m128i*)p, _mm256_extractf128_si256(a, 0));
Sse41::StoreHalf<0>((__m128i*)p + 1, _mm256_extractf128_si256(a, 1));
}
}
#endif//SIMD_SAVX2_ENABLE
#ifdef SIMD_NEON_ENABLE
namespace Neon
{
template <bool align> SIMD_INLINE void Store(uint8_t * p, uint8x16_t a);
template <> SIMD_INLINE void Store<false>(uint8_t * p, uint8x16_t a)
{
vst1q_u8(p, a);
}
template <> SIMD_INLINE void Store<true>(uint8_t * p, uint8x16_t a)
{
#if defined(__GNUC__)
uint8_t * _p = (uint8_t *)__builtin_assume_aligned(p, 16);
vst1q_u8(_p, a);
#elif defined(_MSC_VER)
vst1q_u8_ex(p, a, 128);
#else
vst1q_u8(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store(int8_t* p, int8x16_t a)
{
Store<align>((uint8_t*)p, vreinterpretq_u8_s8(a));
}
template <bool align> SIMD_INLINE void Store(uint8_t * p, uint8x8_t a);
template <> SIMD_INLINE void Store<false>(uint8_t * p, uint8x8_t a)
{
vst1_u8(p, a);
}
template <> SIMD_INLINE void Store<true>(uint8_t * p, uint8x8_t a)
{
#if defined(__GNUC__)
uint8_t * _p = (uint8_t *)__builtin_assume_aligned(p, 8);
vst1_u8(_p, a);
#elif defined(_MSC_VER)
vst1_u8_ex(p, a, 64);
#else
vst1_u8(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store(uint16_t * p, uint16x8_t a)
{
Store<align>((uint8_t*)p, vreinterpretq_u8_u16(a));
}
template <bool align> SIMD_INLINE void Store(uint16_t * p, uint16x4_t a)
{
Store<align>((uint8_t*)p, vreinterpret_u8_u16(a));
}
template <bool align> SIMD_INLINE void Store(int16_t * p, int16x8_t a)
{
Store<align>((uint8_t*)p, vreinterpretq_u8_s16(a));
}
template <bool align> SIMD_INLINE void Store(uint32_t * p, uint32x4_t a)
{
Store<align>((uint8_t*)p, vreinterpretq_u8_u32(a));
}
template <bool align> SIMD_INLINE void Store(int32_t * p, int32x4_t a)
{
Store<align>((uint8_t*)p, vreinterpretq_u8_s32(a));
}
template <bool align> SIMD_INLINE void Store2(uint8_t * p, uint8x16x2_t a);
template <> SIMD_INLINE void Store2<false>(uint8_t * p, uint8x16x2_t a)
{
vst2q_u8(p, a);
}
template <> SIMD_INLINE void Store2<true>(uint8_t * p, uint8x16x2_t a)
{
#if defined(__GNUC__)
uint8_t * _p = (uint8_t *)__builtin_assume_aligned(p, 16);
vst2q_u8(_p, a);
#elif defined(_MSC_VER)
vst2q_u8_ex(p, a, 128);
#else
vst2q_u8(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store2(int16_t * p, int16x8x2_t a);
template <> SIMD_INLINE void Store2<false>(int16_t * p, int16x8x2_t a)
{
vst2q_s16(p, a);
}
template <> SIMD_INLINE void Store2<true>(int16_t * p, int16x8x2_t a)
{
#if defined(__GNUC__)
int16_t * _p = (int16_t *)__builtin_assume_aligned(p, 16);
vst2q_s16(_p, a);
#elif defined(_MSC_VER)
vst2q_s16_ex(p, a, 128);
#else
vst2q_s16(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store2(uint8_t * p, uint8x8x2_t a);
template <> SIMD_INLINE void Store2<false>(uint8_t * p, uint8x8x2_t a)
{
vst2_u8(p, a);
}
template <> SIMD_INLINE void Store2<true>(uint8_t * p, uint8x8x2_t a)
{
#if defined(__GNUC__)
uint8_t * _p = (uint8_t *)__builtin_assume_aligned(p, 8);
vst2_u8(_p, a);
#elif defined(_MSC_VER)
vst2_u8_ex(p, a, 64);
#else
vst2_u8(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store3(uint8_t * p, uint8x16x3_t a);
template <> SIMD_INLINE void Store3<false>(uint8_t * p, uint8x16x3_t a)
{
vst3q_u8(p, a);
}
template <> SIMD_INLINE void Store3<true>(uint8_t * p, uint8x16x3_t a)
{
#if defined(__GNUC__)
uint8_t * _p = (uint8_t *)__builtin_assume_aligned(p, 16);
vst3q_u8(_p, a);
#elif defined(_MSC_VER)
vst3q_u8_ex(p, a, 128);
#else
vst3q_u8(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store3(uint8_t * p, uint8x8x3_t a);
template <> SIMD_INLINE void Store3<false>(uint8_t * p, uint8x8x3_t a)
{
vst3_u8(p, a);
}
template <> SIMD_INLINE void Store3<true>(uint8_t * p, uint8x8x3_t a)
{
#if defined(__GNUC__)
uint8_t * _p = (uint8_t *)__builtin_assume_aligned(p, 8);
vst3_u8(_p, a);
#elif defined(_MSC_VER)
vst3_u8_ex(p, a, 64);
#else
vst3_u8(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store4(uint8_t * p, uint8x16x4_t a);
template <> SIMD_INLINE void Store4<false>(uint8_t * p, uint8x16x4_t a)
{
vst4q_u8(p, a);
}
template <> SIMD_INLINE void Store4<true>(uint8_t * p, uint8x16x4_t a)
{
#if defined(__GNUC__)
uint8_t * _p = (uint8_t *)__builtin_assume_aligned(p, 16);
vst4q_u8(_p, a);
#elif defined(_MSC_VER)
vst4q_u8_ex(p, a, 128);
#else
vst4q_u8(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store4(uint8_t * p, uint8x8x4_t a);
template <> SIMD_INLINE void Store4<false>(uint8_t * p, uint8x8x4_t a)
{
vst4_u8(p, a);
}
template <> SIMD_INLINE void Store4<true>(uint8_t * p, uint8x8x4_t a)
{
#if defined(__GNUC__)
uint8_t * _p = (uint8_t *)__builtin_assume_aligned(p, 8);
vst4_u8(_p, a);
#elif defined(_MSC_VER)
vst4_u8_ex(p, a, 64);
#else
vst4_u8(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store(float * p, float32x4_t a);
template <> SIMD_INLINE void Store<false>(float * p, float32x4_t a)
{
vst1q_f32(p, a);
}
template <> SIMD_INLINE void Store<true>(float * p, float32x4_t a)
{
#if defined(__GNUC__)
float * _p = (float *)__builtin_assume_aligned(p, 16);
vst1q_f32(_p, a);
#elif defined(_MSC_VER)
vst1q_f32_ex(p, a, 128);
#else
vst1q_f32(p, a);
#endif
}
SIMD_INLINE void Store(float* ptr, float32x4_t val, size_t size)
{
SIMD_ALIGNED(16) float buf[F];
Store<true>(buf, val);
for (size_t i = 0; i < size; ++i)
ptr[i] = buf[i];
}
template <bool align> SIMD_INLINE void Store(float * p, float32x2_t a);
template <> SIMD_INLINE void Store<false>(float * p, float32x2_t a)
{
vst1_f32(p, a);
}
template <> SIMD_INLINE void Store<true>(float * p, float32x2_t a)
{
#if defined(__GNUC__)
float * _p = (float *)__builtin_assume_aligned(p, 8);
vst1_f32(_p, a);
#elif defined(_MSC_VER)
vst1_f32_ex(p, a, 64);
#else
vst1_f32(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store2(float * p, float32x4x2_t a);
template <> SIMD_INLINE void Store2<false>(float * p, float32x4x2_t a)
{
vst2q_f32(p, a);
}
template <> SIMD_INLINE void Store2<true>(float * p, float32x4x2_t a)
{
#if defined(__GNUC__)
float * _p = (float *)__builtin_assume_aligned(p, 16);
vst2q_f32(_p, a);
#elif defined(_MSC_VER)
vst2q_f32_ex(p, a, 128);
#else
vst2q_f32(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store3(float * p, float32x4x3_t a);
template <> SIMD_INLINE void Store3<false>(float * p, float32x4x3_t a)
{
vst3q_f32(p, a);
}
template <> SIMD_INLINE void Store3<true>(float * p, float32x4x3_t a)
{
#if defined(__GNUC__)
float * _p = (float *)__builtin_assume_aligned(p, 16);
vst3q_f32(_p, a);
#elif defined(_MSC_VER)
vst3q_f32_ex(p, a, 128);
#else
vst3q_f32(p, a);
#endif
}
template <bool align> SIMD_INLINE void Store4(float * p, float32x4x4_t a);
template <> SIMD_INLINE void Store4<false>(float * p, float32x4x4_t a)
{
vst4q_f32(p, a);
}
template <> SIMD_INLINE void Store4<true>(float * p, float32x4x4_t a)
{
#if defined(__GNUC__)
float * _p = (float *)__builtin_assume_aligned(p, 16);
vst4q_f32(_p, a);
#elif defined(_MSC_VER)
vst4q_f32_ex(p, a, 128);
#else
vst4q_f32(p, a);
#endif
}
template <bool align> SIMD_INLINE void StoreMasked(float * p, float32x4_t value, uint32x4_t mask)
{
float32x4_t old = Load<align>(p);
Store<align>(p, vbslq_f32(mask, value, old));
}
}
#endif//SIMD_NEON_ENABLE
}
#endif//__SimdStore_h__
|
