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
|
// SIMD helper
// optimze based on technolegy double, float and integer (32) SIMD instructions
// writen by Martin Steinegger
#ifndef SIMD_H
#define SIMD_H
#include <cstdlib>
#include <limits>
#include <algorithm>
#include <iostream>
#define AVX512_ALIGN_DOUBLE 64
#define AVX512_VECSIZE_DOUBLE 8
#define AVX512_ALIGN_FLOAT 64
#define AVX512_VECSIZE_FLOAT 16
#define AVX512_ALIGN_INT 64
#define AVX512_VECSIZE_INT 16
#define AVX_ALIGN_DOUBLE 32
#define AVX_VECSIZE_DOUBLE 4
#define AVX_ALIGN_FLOAT 32
#define AVX_VECSIZE_FLOAT 8
#define AVX2_ALIGN_INT 32
#define AVX2_VECSIZE_INT 8
#define SSE_ALIGN_DOUBLE 16
#define SSE_VECSIZE_DOUBLE 2
#define SSE_ALIGN_FLOAT 16
#define SSE_VECSIZE_FLOAT 4
#define SSE_ALIGN_INT 16
#define SSE_VECSIZE_INT 4
#define MAX_ALIGN_DOUBLE AVX512_ALIGN_DOUBLE
#define MAX_VECSIZE_DOUBLE AVX512_VECSIZE_DOUBLE
#define MAX_ALIGN_FLOAT AVX512_ALIGN_FLOAT
#define MAX_VECSIZE_FLOAT AVX512_VECSIZE_FLOAT
#define MAX_ALIGN_INT AVX512_ALIGN_INT
#define MAX_VECSIZE_INT AVX512_VECSIZE_INT
#define SIMDE_ENABLE_NATIVE_ALIASES
#include <simde/simde-features.h>
// FIXME: Finish AVX512 implementation
//#if defined(SIMDE_X86_AVX512F_NATIVE) && defined(SIMDE_X86_AVX512BW_NATIVE)
//#define AVX512
//#endif
#if defined(AVX512) || defined(SIMDE_X86_AVX2_NATIVE)
#define AVX2
#endif
#ifdef AVX512
#include <simde/x86/avx512f.h>
#include <simde/x86/avx512bw.h>
// double support
#ifndef SIMD_DOUBLE
#define SIMD_DOUBLE
#define ALIGN_DOUBLE AVX512_ALIGN_DOUBLE
#define VECSIZE_DOUBLE AVX512_VECSIZE_DOUBLE
typedef __m512d simd_double;
#define simdf64_add(x,y) _mm512_add_pd(x,y)
#define simdf64_sub(x,y) _mm512_sub_pd(x,y)
#define simdf64_mul(x,y) _mm512_mul_pd(x,y)
#define simdf64_div(x,y) _mm512_div_pd(x,y)
#define simdf64_max(x,y) _mm512_max_pd(x,y)
#define simdf64_load(x) _mm512_load_pd(x)
#define simdf64_store(x,y) _mm512_store_pd(x,y)
#define simdf64_set(x) _mm512_set1_pd(x)
#define simdf64_setzero(x) _mm512_setzero_pd()
#define simdf64_gt(x,y) _mm512_cmpnle_pd_mask(x,y)
#define simdf64_lt(x,y) _mm512_cmplt_pd_mask(x,y)
#define simdf64_or(x,y) _mm512_or_si512(x,y)
#define simdf64_and(x,y) _mm512_and_si512 (x,y)
#define simdf64_andnot(x,y) _mm512_andnot_si512(x,y)
#define simdf64_xor(x,y) _mm512_xor_si512(x,y)
#endif //SIMD_DOUBLE
// float support
#ifndef SIMD_FLOAT
#define SIMD_FLOAT
#define ALIGN_FLOAT AVX512_ALIGN_FLOAT
#define VECSIZE_FLOAT AVX512_VECSIZE_FLOAT
typedef __m512 simd_float;
#define simdf32_add(x,y) _mm512_add_ps(x,y)
#define simdf32_sub(x,y) _mm512_sub_ps(x,y)
#define simdf32_mul(x,y) _mm512_mul_ps(x,y)
#define simdf32_div(x,y) _mm512_div_ps(x,y)
#define simdf32_rcp(x) _mm512_rcp_ps(x)
#define simdf32_max(x,y) _mm512_max_ps(x,y)
#define simdf32_min(x,y) _mm512_min_ps(x,y)
#define simdf32_load(x) _mm512_load_ps(x)
#define simdf32_store(x,y) _mm512_store_ps(x,y)
#define simdf32_set(x) _mm512_set1_ps(x)
#define simdf32_setzero(x) _mm512_setzero_ps()
#define simdf32_gt(x,y) _mm512_cmpnle_ps_mask(x,y)
#define simdf32_eq(x,y) _mm512_cmpeq_ps_mask(x,y)
#define simdf32_lt(x,y) _mm512_cmplt_ps_mask(x,y)
#define simdf32_or(x,y) _mm512_or_si512(x,y)
#define simdf32_and(x,y) _mm512_and_si512(x,y)
#define simdf32_andnot(x,y) _mm512_andnot_si512(x,y)
#define simdf32_xor(x,y) _mm512_xor_si512(x,y)
#define simdf32_f2i(x) _mm512_cvtps_epi32(x) // convert s.p. float to integer
#define simdf_f2icast(x) _mm512_castps_si512(x)
#endif //SIMD_FLOAT
// integer support
#ifndef SIMD_INT
#define SIMD_INT
#define ALIGN_INT AVX512_ALIGN_INT
#define VECSIZE_INT AVX512_VECSIZE_INT
typedef __m512i simd_int;
#define simdi32_add(x,y) _mm512_add_epi32(x,y)
#define simdi16_add(x,y) _mm512_add_epi16(x,y)
#define simdi16_adds(x,y) _mm512_adds_epi16(x,y)
#define simdui8_adds(x,y) _mm512_adds_epu8()
#define simdi32_sub(x,y) _mm512_sub_epi32(x,y)
#define simdui8_subs(x,y) _mm512_subs_epu8()
#define simdi32_mul(x,y) _mm512_mullo_epi32(x,y)
#define simdui8_max(x,y) _mm512_max_epu8()
#define simdi16_max(x,y) _mm512_max_epi16(x,y)
#define simdi32_max(x,y) _mm512_max_epi32(x,y)
#define simdi_load(x) _mm512_load_si512(x)
#define simdi_streamload(x) _mm512_stream_load_si512(x)
#define simdi_store(x,y) _mm512_store_si512(x,y)
#define simdi_storeu(x,y) _mm512_storeu_si512(x,y)
#define simdi32_set(x) _mm512_set1_epi32(x)
#define simdi16_set(x) _mm512_set1_epi16(x)
#define simdi8_set(x) _mm512_set1_epi8(x)
#define simdi32_shuffle(x,y) _mm512_shuffle_epi32(x,y)
#define simdi16_shuffle(x,y) NOT_YET_IMP(x,y)
#define simdi8_shuffle(x,y) _mm512_shuffle_epi8(x,y)
#define simdi_setzero() _mm512_setzero_si512()
#define simdi32_gt(x,y) _mm512_cmpgt_epi32(x,y)
#define simdi8_gt(x,y) NOT_YET_IMP()
#define simdi16_gt(x,y) NOT_YET_IMP()
#define simdi8_eq(x,y) NOT_YET_IMP()
#define simdi32_lt(x,y) NOT_YET_IMP()
#define simdi16_lt(x,y) NOT_YET_IMP()
#define simdi8_lt(x,y) NOT_YET_IMP()
#define simdi_or(x,y) _mm512_or_si512(x,y)
#define simdi_and(x,y) _mm512_and_si512(x,y)
#define simdi_andnot(x,y) _mm512_andnot_si512(x,y)
#define simdi_xor(x,y) _mm512_xor_si512(x,y)
#define simdi8_shiftl(x,y) NOT_YET_IMP()
#define simdi8_shiftr(x,y) NOT_YET_IMP()
#define simdi8_movemask(x) NOT_YET_IMP()
#define simdi16_extract(x,y) NOT_YET_IMP()
#define simdi16_slli(x,y) _mm512_slli_epi16(x,y) // shift integers in a left by y
#define simdi16_srli(x,y) _mm512_srli_epi16(x,y) // shift integers in a right by y
#define simdi32_slli(x,y) _mm512_slli_epi32(x,y) // shift integers in a left by y
#define simdi32_srli(x,y) _mm512_srli_epi32(x,y) // shift integers in a right by y
#define simdi32_i2f(x) _mm512_cvtepi32_ps(x) // convert integer to s.p. float
#define simdi_i2fcast(x) _mm512_castsi512_ps(x)
#endif //SIMD_INT
#endif //AVX512_SUPPORT
#ifdef AVX2
#include <simde/x86/avx2.h>
// integer support (usable with AVX2)
#ifndef SIMD_INT
#define SIMD_INT
#define ALIGN_INT AVX2_ALIGN_INT
#define VECSIZE_INT AVX2_VECSIZE_INT
uint16_t simd_hmax16_sse(const __m128i buffer);
uint8_t simd_hmax8_sse(const __m128i buffer);
inline uint16_t simd_hmax16_avx(const __m256i buffer) {
const __m128i abcd = _mm256_castsi256_si128(buffer);
const uint16_t first = simd_hmax16_sse(abcd);
const __m128i efgh = _mm256_extracti128_si256(buffer, 1);
const uint16_t second = simd_hmax16_sse(efgh);
return std::max(first, second);
}
inline uint8_t simd_hmax8_avx(const __m256i buffer) {
const __m128i abcd = _mm256_castsi256_si128(buffer);
const uint8_t first = simd_hmax8_sse(abcd);
const __m128i efgh = _mm256_extracti128_si256(buffer, 1);
const uint8_t second = simd_hmax8_sse(efgh);
return std::max(first, second);
}
template <unsigned int N>
inline __m256i _mm256_shift_left(__m256i a) {
__m256i mask = _mm256_permute2x128_si256(a, a, _MM_SHUFFLE(0,0,3,0) );
return _mm256_alignr_epi8(a,mask,16-N);
}
inline unsigned short extract_epi16(__m256i v, int pos) {
switch(pos){
case 0: return _mm256_extract_epi16(v, 0);
case 1: return _mm256_extract_epi16(v, 1);
case 2: return _mm256_extract_epi16(v, 2);
case 3: return _mm256_extract_epi16(v, 3);
case 4: return _mm256_extract_epi16(v, 4);
case 5: return _mm256_extract_epi16(v, 5);
case 6: return _mm256_extract_epi16(v, 6);
case 7: return _mm256_extract_epi16(v, 7);
case 8: return _mm256_extract_epi16(v, 8);
case 9: return _mm256_extract_epi16(v, 9);
case 10: return _mm256_extract_epi16(v, 10);
case 11: return _mm256_extract_epi16(v, 11);
case 12: return _mm256_extract_epi16(v, 12);
case 13: return _mm256_extract_epi16(v, 13);
case 14: return _mm256_extract_epi16(v, 14);
case 15: return _mm256_extract_epi16(v, 15);
}
return 0;
}
typedef __m256i simd_int;
#define simdi32_add(x,y) _mm256_add_epi32(x,y)
#define simdi16_add(x,y) _mm256_add_epi16(x,y)
#define simdi16_adds(x,y) _mm256_adds_epi16(x,y)
#define simdui8_adds(x,y) _mm256_adds_epu8(x,y)
#define simdi32_sub(x,y) _mm256_sub_epi32(x,y)
#define simdui16_subs(x,y) _mm256_subs_epu16(x,y)
#define simdui8_subs(x,y) _mm256_subs_epu8(x,y)
#define simdi32_mul(x,y) _mm256_mullo_epi32(x,y)
#define simdi32_max(x,y) _mm256_max_epi32(x,y)
#define simdi16_max(x,y) _mm256_max_epi16(x,y)
#define simdi16_hmax(x) simd_hmax16_avx(x)
#define simdui8_max(x,y) _mm256_max_epu8(x,y)
#define simdi8_hmax(x) simd_hmax8_avx(x)
#define simdi_load(x) _mm256_load_si256(x)
#define simdi_loadu(x) _mm256_loadu_si256(x)
#define simdi_streamload(x) _mm256_stream_load_si256(x)
#define simdi_store(x,y) _mm256_store_si256(x,y)
#define simdi_storeu(x,y) _mm256_storeu_si256(x,y)
#define simdi32_set(x) _mm256_set1_epi32(x)
#define simdi16_set(x) _mm256_set1_epi16(x)
#define simdi8_set(x) _mm256_set1_epi8(x)
#define simdi32_shuffle(x,y) _mm256_shuffle_epi32(x,y)
#define simdi16_shuffle(x,y) _mm256_shuffle_epi16(x,y)
#define simdi8_shuffle(x,y) _mm256_shuffle_epi8(x,y)
#define simdi_setzero() _mm256_setzero_si256()
#define simdi32_gt(x,y) _mm256_cmpgt_epi32(x,y)
#define simdi8_gt(x,y) _mm256_cmpgt_epi8(x,y)
#define simdi16_gt(x,y) _mm256_cmpgt_epi16(x,y)
#define simdi8_eq(x,y) _mm256_cmpeq_epi8(x,y)
#define simdi16_eq(x,y) _mm256_cmpeq_epi16(x,y)
#define simdi32_eq(x,y) _mm256_cmpeq_epi32(x,y)
#define simdi32_lt(x,y) _mm256_cmpgt_epi32(y,x) // inverse
#define simdi16_lt(x,y) _mm256_cmpgt_epi16(y,x) // inverse
#define simdi8_lt(x,y) _mm256_cmpgt_epi8(y,x)
#define simdi_or(x,y) _mm256_or_si256(x,y)
#define simdi_and(x,y) _mm256_and_si256(x,y)
#define simdi_andnot(x,y) _mm256_andnot_si256(x,y)
#define simdi_xor(x,y) _mm256_xor_si256(x,y)
#define simdi8_shiftl(x,y) _mm256_shift_left<y>(x)
//TODO fix like shift_left
#define simdi8_shiftr(x,y) _mm256_srli_si256(x,y)
#define SIMD_MOVEMASK_MAX 0xffffffff
#define simdi8_movemask(x) _mm256_movemask_epi8(x)
#define simdi16_extract(x,y) extract_epi16(x,y)
#define simdi16_slli(x,y) _mm256_slli_epi16(x,y) // shift integers in a left by y
#define simdi16_srli(x,y) _mm256_srli_epi16(x,y) // shift integers in a right by y
#define simdi32_slli(x,y) _mm256_slli_epi32(x,y) // shift integers in a left by y
#define simdi32_srli(x,y) _mm256_srli_epi32(x,y) // shift integers in a right by y
#define simdi32_i2f(x) _mm256_cvtepi32_ps(x) // convert integer to s.p. float
#define simdi_i2fcast(x) _mm256_castsi256_ps(x)
#endif
#include <simde/x86/avx.h>
// double support (usable with AVX1)
#ifndef SIMD_DOUBLE
#define SIMD_DOUBLE
#define ALIGN_DOUBLE AVX_ALIGN_DOUBLE
#define VECSIZE_DOUBLE AVX_VECSIZE_DOUBLE
typedef __m256d simd_double;
#define simdf64_add(x,y) _mm256_add_pd(x,y)
#define simdf64_sub(x,y) _mm256_sub_pd(x,y)
#define simdf64_mul(x,y) _mm256_mul_pd(x,y)
#define simdf64_div(x,y) _mm256_div_pd(x,y)
#define simdf64_max(x,y) _mm256_max_pd(x,y)
#define simdf64_load(x) _mm256_load_pd(x)
#define simdf64_store(x,y) _mm256_store_pd(x,y)
#define simdf64_set(x) _mm256_set1_pd(x)
#define simdf64_setzero(x) _mm256_setzero_pd()
#define simdf64_gt(x,y) _mm256_cmp_pd(x,y,_CMP_GT_OS)
#define simdf64_lt(x,y) _mm256_cmp_pd(x,y,_CMP_LT_OS)
#define simdf64_or(x,y) _mm256_or_pd(x,y)
#define simdf64_and(x,y) _mm256_and_pd(x,y)
#define simdf64_andnot(x,y) _mm256_andnot_pd(x,y)
#define simdf64_xor(x,y) _mm256_xor_pd(x,y)
#endif //SIMD_DOUBLE
// float support (usable with AVX1)
#ifndef SIMD_FLOAT
#define SIMD_FLOAT
#define ALIGN_FLOAT AVX_ALIGN_FLOAT
#define VECSIZE_FLOAT AVX_VECSIZE_FLOAT
typedef __m256 simd_float;
#define simdf32_add(x,y) _mm256_add_ps(x,y)
#define simdf32_sub(x,y) _mm256_sub_ps(x,y)
#define simdf32_mul(x,y) _mm256_mul_ps(x,y)
#define simdf32_div(x,y) _mm256_div_ps(x,y)
#define simdf32_rcp(x) _mm256_rcp_ps(x)
#define simdf32_max(x,y) _mm256_max_ps(x,y)
#define simdf32_min(x,y) _mm256_min_ps(x,y)
#define simdf32_load(x) _mm256_load_ps(x)
#define simdf32_store(x,y) _mm256_store_ps(x,y)
#define simdf32_set(x) _mm256_set1_ps(x)
#define simdf32_setzero(x) _mm256_setzero_ps()
#define simdf32_gt(x,y) _mm256_cmp_ps(x,y,_CMP_GT_OS)
#define simdf32_eq(x,y) _mm256_cmp_ps(x,y,_CMP_EQ_OS)
#define simdf32_lt(x,y) _mm256_cmp_ps(x,y,_CMP_LT_OS)
#define simdf32_or(x,y) _mm256_or_ps(x,y)
#define simdf32_and(x,y) _mm256_and_ps(x,y)
#define simdf32_andnot(x,y) _mm256_andnot_ps(x,y)
#define simdf32_xor(x,y) _mm256_xor_ps(x,y)
#define simdf32_f2i(x) _mm256_cvtps_epi32(x) // convert s.p. float to integer
#define simdf_f2icast(x) _mm256_castps_si256(x) // compile time cast
#endif
#endif
#include <simde/x86/sse4.1.h>
inline uint16_t simd_hmax16_sse(const __m128i buffer) {
__m128i tmp1 = _mm_subs_epu16(_mm_set1_epi16((short)65535), buffer);
__m128i tmp3 = _mm_minpos_epu16(tmp1);
return (65535 - _mm_cvtsi128_si32(tmp3));
}
inline uint8_t simd_hmax8_sse(const __m128i buffer) {
__m128i tmp1 = _mm_subs_epu8(_mm_set1_epi8((char)255), buffer);
__m128i tmp2 = _mm_min_epu8(tmp1, _mm_srli_epi16(tmp1, 8));
__m128i tmp3 = _mm_minpos_epu16(tmp2);
return (int8_t)(255 -(int8_t) _mm_cvtsi128_si32(tmp3));
}
// double support
#ifndef SIMD_DOUBLE
#define SIMD_DOUBLE
#define ALIGN_DOUBLE SSE_ALIGN_DOUBLE
#define VECSIZE_DOUBLE SSE_VECSIZE_DOUBLE
typedef __m128d simd_double;
#define simdf64_add(x,y) _mm_add_pd(x,y)
#define simdf64_sub(x,y) _mm_sub_pd(x,y)
#define simdf64_mul(x,y) _mm_mul_pd(x,y)
#define simdf64_div(x,y) _mm_div_pd(x,y)
#define simdf64_max(x,y) _mm_max_pd(x,y)
#define simdf64_load(x) _mm_load_pd(x)
#define simdf64_store(x,y) _mm_store_pd(x,y)
#define simdf64_set(x) _mm_set1_pd(x)
#define simdf64_setzero(x) _mm_setzero_pd()
#define simdf64_gt(x,y) _mm_cmpgt_pd(x,y)
#define simdf64_lt(x,y) _mm_cmplt_pd(x,y)
#define simdf64_or(x,y) _mm_or_pd(x,y)
#define simdf64_and(x,y) _mm_and_pd(x,y)
#define simdf64_andnot(x,y) _mm_andnot_pd(x,y)
#define simdf64_xor(x,y) _mm_xor_pd(x,y)
#endif //SIMD_DOUBLE
// float support
#ifndef SIMD_FLOAT
#define SIMD_FLOAT
#define ALIGN_FLOAT SSE_ALIGN_FLOAT
#define VECSIZE_FLOAT SSE_VECSIZE_FLOAT
typedef __m128 simd_float;
#define simdf32_add(x,y) _mm_add_ps(x,y)
#define simdf32_sub(x,y) _mm_sub_ps(x,y)
#define simdf32_mul(x,y) _mm_mul_ps(x,y)
#define simdf32_div(x,y) _mm_div_ps(x,y)
#define simdf32_rcp(x) _mm_rcp_ps(x)
#define simdf32_max(x,y) _mm_max_ps(x,y)
#define simdf32_min(x,y) _mm_min_ps(x,y)
#define simdf32_load(x) _mm_load_ps(x)
#define simdf32_store(x,y) _mm_store_ps(x,y)
#define simdf32_set(x) _mm_set1_ps(x)
#define simdf32_setzero(x) _mm_setzero_ps()
#define simdf32_gt(x,y) _mm_cmpgt_ps(x,y)
#define simdf32_eq(x,y) _mm_cmpeq_ps(x,y)
#define simdf32_lt(x,y) _mm_cmplt_ps(x,y)
#define simdf32_or(x,y) _mm_or_ps(x,y)
#define simdf32_and(x,y) _mm_and_ps(x,y)
#define simdf32_andnot(x,y) _mm_andnot_ps(x,y)
#define simdf32_xor(x,y) _mm_xor_ps(x,y)
#define simdf32_f2i(x) _mm_cvtps_epi32(x) // convert s.p. float to integer
#define simdf_f2icast(x) _mm_castps_si128(x) // compile time cast
#endif //SIMD_FLOAT
// integer support
#ifndef SIMD_INT
#define SIMD_INT
inline unsigned short extract_epi16(__m128i v, int pos) {
switch(pos){
case 0: return _mm_extract_epi16(v, 0);
case 1: return _mm_extract_epi16(v, 1);
case 2: return _mm_extract_epi16(v, 2);
case 3: return _mm_extract_epi16(v, 3);
case 4: return _mm_extract_epi16(v, 4);
case 5: return _mm_extract_epi16(v, 5);
case 6: return _mm_extract_epi16(v, 6);
case 7: return _mm_extract_epi16(v, 7);
}
return 0;
}
#define ALIGN_INT SSE_ALIGN_INT
#define VECSIZE_INT SSE_VECSIZE_INT
typedef __m128i simd_int;
#define simdi32_add(x,y) _mm_add_epi32(x,y)
#define simdi16_add(x,y) _mm_add_epi16(x,y)
#define simdi16_adds(x,y) _mm_adds_epi16(x,y)
#define simdui8_adds(x,y) _mm_adds_epu8(x,y)
#define simdi32_sub(x,y) _mm_sub_epi32(x,y)
#define simdui16_subs(x,y) _mm_subs_epu16(x,y)
#define simdui8_subs(x,y) _mm_subs_epu8(x,y)
#define simdi32_mul(x,y) _mm_mullo_epi32(x,y) // SSE4.1
#define simdi32_max(x,y) _mm_max_epi32(x,y) // SSE4.1
#define simdi16_max(x,y) _mm_max_epi16(x,y)
#define simdi16_hmax(x) simd_hmax16_sse(x)
#define simdui8_max(x,y) _mm_max_epu8(x,y)
#define simdi8_hmax(x) simd_hmax8_sse(x)
#define simdi_load(x) _mm_load_si128(x)
#define simdi_loadu(x) _mm_loadu_si128(x)
#define simdi_streamload(x) _mm_stream_load_si128(x)
#define simdi_storeu(x,y) _mm_storeu_si128(x,y)
#define simdi_store(x,y) _mm_store_si128(x,y)
#define simdi32_set(x) _mm_set1_epi32(x)
#define simdi16_set(x) _mm_set1_epi16(x)
#define simdi8_set(x) _mm_set1_epi8(x)
#define simdi32_shuffle(x,y) _mm_shuffle_epi32(x,y)
#define simdi16_shuffle(x,y) _mm_shuffle_epi16(x,y)
#define simdi8_shuffle(x,y) _mm_shuffle_epi8(x,y)
#define simdi_setzero() _mm_setzero_si128()
#define simdi32_gt(x,y) _mm_cmpgt_epi32(x,y)
#define simdi8_gt(x,y) _mm_cmpgt_epi8(x,y)
#define simdi32_eq(x,y) _mm_cmpeq_epi32(x,y)
#define simdi16_eq(x,y) _mm_cmpeq_epi16(x,y)
#define simdi8_eq(x,y) _mm_cmpeq_epi8(x,y)
#define simdi32_lt(x,y) _mm_cmplt_epi32(x,y)
#define simdi16_lt(x,y) _mm_cmplt_epi16(x,y)
#define simdi8_lt(x,y) _mm_cmplt_epi8(x,y)
#define simdi16_gt(x,y) _mm_cmpgt_epi16(x,y)
#define simdi_or(x,y) _mm_or_si128(x,y)
#define simdi_and(x,y) _mm_and_si128(x,y)
#define simdi_andnot(x,y) _mm_andnot_si128(x,y)
#define simdi_xor(x,y) _mm_xor_si128(x,y)
#define simdi8_shiftl(x,y) _mm_slli_si128(x,y)
#define simdi8_shiftr(x,y) _mm_srli_si128(x,y)
#define SIMD_MOVEMASK_MAX 0xffff
#define simdi8_movemask(x) _mm_movemask_epi8(x)
#define simdi16_extract(x,y) extract_epi16(x,y)
#define simdi16_slli(x,y) _mm_slli_epi16(x,y) // shift integers in a left by y
#define simdi16_srli(x,y) _mm_srli_epi16(x,y) // shift integers in a right by y
#define simdi32_slli(x,y) _mm_slli_epi32(x,y) // shift integers in a left by y
#define simdi32_srli(x,y) _mm_srli_epi32(x,y) // shift integers in a right by y
#define simdi32_i2f(x) _mm_cvtepi32_ps(x) // convert integer to s.p. float
#define simdi_i2fcast(x) _mm_castsi128_ps(x)
#endif //SIMD_INT
inline void *mem_align(size_t boundary, size_t size) {
void *pointer;
if (posix_memalign(&pointer, boundary, size) != 0) {
#define MEM_ALIGN_ERROR "mem_align could not allocate memory.\n"
fwrite(MEM_ALIGN_ERROR, sizeof(MEM_ALIGN_ERROR), 1, stderr);
#undef MEM_ALIGN_ERROR
exit(3);
}
return pointer;
}
#ifdef SIMD_FLOAT
inline simd_float * malloc_simd_float(const size_t size) {
return (simd_float *) mem_align(ALIGN_FLOAT, size);
}
#endif
#ifdef SIMD_DOUBLE
inline simd_double * malloc_simd_double(const size_t size) {
return (simd_double *) mem_align(ALIGN_DOUBLE, size);
}
#endif
#ifdef SIMD_INT
inline simd_int * malloc_simd_int(const size_t size) {
return (simd_int *) mem_align(ALIGN_INT, size);
}
#endif
#endif //SIMD_H
|
