]>
Commit | Line | Data |
---|---|---|
a945c346 | 1 | /* Copyright (C) 2008-2024 Free Software Foundation, Inc. |
95879c72 L |
2 | |
3 | This file is part of GCC. | |
4 | ||
5 | GCC is free software; you can redistribute it and/or modify | |
6 | it under the terms of the GNU General Public License as published by | |
748086b7 | 7 | the Free Software Foundation; either version 3, or (at your option) |
95879c72 L |
8 | any later version. |
9 | ||
10 | GCC is distributed in the hope that it will be useful, | |
11 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | GNU General Public License for more details. | |
14 | ||
748086b7 JJ |
15 | Under Section 7 of GPL version 3, you are granted additional |
16 | permissions described in the GCC Runtime Library Exception, version | |
17 | 3.1, as published by the Free Software Foundation. | |
18 | ||
19 | You should have received a copy of the GNU General Public License and | |
20 | a copy of the GCC Runtime Library Exception along with this program; | |
21 | see the files COPYING3 and COPYING.RUNTIME respectively. If not, see | |
22 | <http://www.gnu.org/licenses/>. */ | |
95879c72 L |
23 | |
24 | /* Implemented from the specification included in the Intel C++ Compiler | |
25 | User Guide and Reference, version 11.0. */ | |
26 | ||
37fe763d UB |
27 | #ifndef _IMMINTRIN_H_INCLUDED |
28 | # error "Never use <avxintrin.h> directly; include <immintrin.h> instead." | |
29 | #endif | |
95879c72 | 30 | |
97db2bf7 ST |
31 | #ifndef _AVXINTRIN_H_INCLUDED |
32 | #define _AVXINTRIN_H_INCLUDED | |
33 | ||
34 | #ifndef __AVX__ | |
35 | #pragma GCC push_options | |
36 | #pragma GCC target("avx") | |
37 | #define __DISABLE_AVX__ | |
38 | #endif /* __AVX__ */ | |
39 | ||
95879c72 L |
40 | /* Internal data types for implementing the intrinsics. */ |
41 | typedef double __v4df __attribute__ ((__vector_size__ (32))); | |
42 | typedef float __v8sf __attribute__ ((__vector_size__ (32))); | |
43 | typedef long long __v4di __attribute__ ((__vector_size__ (32))); | |
2069d6fc | 44 | typedef unsigned long long __v4du __attribute__ ((__vector_size__ (32))); |
95879c72 | 45 | typedef int __v8si __attribute__ ((__vector_size__ (32))); |
2069d6fc | 46 | typedef unsigned int __v8su __attribute__ ((__vector_size__ (32))); |
95879c72 | 47 | typedef short __v16hi __attribute__ ((__vector_size__ (32))); |
2069d6fc | 48 | typedef unsigned short __v16hu __attribute__ ((__vector_size__ (32))); |
95879c72 | 49 | typedef char __v32qi __attribute__ ((__vector_size__ (32))); |
b245befc | 50 | typedef signed char __v32qs __attribute__ ((__vector_size__ (32))); |
2069d6fc | 51 | typedef unsigned char __v32qu __attribute__ ((__vector_size__ (32))); |
95879c72 L |
52 | |
53 | /* The Intel API is flexible enough that we must allow aliasing with other | |
54 | vector types, and their scalar components. */ | |
55 | typedef float __m256 __attribute__ ((__vector_size__ (32), | |
56 | __may_alias__)); | |
57 | typedef long long __m256i __attribute__ ((__vector_size__ (32), | |
58 | __may_alias__)); | |
59 | typedef double __m256d __attribute__ ((__vector_size__ (32), | |
60 | __may_alias__)); | |
61 | ||
c6b0037d MG |
62 | /* Unaligned version of the same types. */ |
63 | typedef float __m256_u __attribute__ ((__vector_size__ (32), | |
64 | __may_alias__, | |
65 | __aligned__ (1))); | |
66 | typedef long long __m256i_u __attribute__ ((__vector_size__ (32), | |
67 | __may_alias__, | |
68 | __aligned__ (1))); | |
69 | typedef double __m256d_u __attribute__ ((__vector_size__ (32), | |
70 | __may_alias__, | |
71 | __aligned__ (1))); | |
72 | ||
95879c72 L |
73 | /* Compare predicates for scalar and packed compare intrinsics. */ |
74 | ||
75 | /* Equal (ordered, non-signaling) */ | |
76 | #define _CMP_EQ_OQ 0x00 | |
77 | /* Less-than (ordered, signaling) */ | |
78 | #define _CMP_LT_OS 0x01 | |
79 | /* Less-than-or-equal (ordered, signaling) */ | |
80 | #define _CMP_LE_OS 0x02 | |
81 | /* Unordered (non-signaling) */ | |
82 | #define _CMP_UNORD_Q 0x03 | |
83 | /* Not-equal (unordered, non-signaling) */ | |
84 | #define _CMP_NEQ_UQ 0x04 | |
85 | /* Not-less-than (unordered, signaling) */ | |
86 | #define _CMP_NLT_US 0x05 | |
87 | /* Not-less-than-or-equal (unordered, signaling) */ | |
88 | #define _CMP_NLE_US 0x06 | |
89 | /* Ordered (nonsignaling) */ | |
90 | #define _CMP_ORD_Q 0x07 | |
91 | /* Equal (unordered, non-signaling) */ | |
92 | #define _CMP_EQ_UQ 0x08 | |
93 | /* Not-greater-than-or-equal (unordered, signaling) */ | |
94 | #define _CMP_NGE_US 0x09 | |
95 | /* Not-greater-than (unordered, signaling) */ | |
96 | #define _CMP_NGT_US 0x0a | |
97 | /* False (ordered, non-signaling) */ | |
98 | #define _CMP_FALSE_OQ 0x0b | |
99 | /* Not-equal (ordered, non-signaling) */ | |
100 | #define _CMP_NEQ_OQ 0x0c | |
101 | /* Greater-than-or-equal (ordered, signaling) */ | |
102 | #define _CMP_GE_OS 0x0d | |
103 | /* Greater-than (ordered, signaling) */ | |
104 | #define _CMP_GT_OS 0x0e | |
105 | /* True (unordered, non-signaling) */ | |
106 | #define _CMP_TRUE_UQ 0x0f | |
107 | /* Equal (ordered, signaling) */ | |
108 | #define _CMP_EQ_OS 0x10 | |
109 | /* Less-than (ordered, non-signaling) */ | |
110 | #define _CMP_LT_OQ 0x11 | |
111 | /* Less-than-or-equal (ordered, non-signaling) */ | |
112 | #define _CMP_LE_OQ 0x12 | |
113 | /* Unordered (signaling) */ | |
114 | #define _CMP_UNORD_S 0x13 | |
115 | /* Not-equal (unordered, signaling) */ | |
116 | #define _CMP_NEQ_US 0x14 | |
117 | /* Not-less-than (unordered, non-signaling) */ | |
118 | #define _CMP_NLT_UQ 0x15 | |
119 | /* Not-less-than-or-equal (unordered, non-signaling) */ | |
120 | #define _CMP_NLE_UQ 0x16 | |
121 | /* Ordered (signaling) */ | |
122 | #define _CMP_ORD_S 0x17 | |
123 | /* Equal (unordered, signaling) */ | |
124 | #define _CMP_EQ_US 0x18 | |
125 | /* Not-greater-than-or-equal (unordered, non-signaling) */ | |
126 | #define _CMP_NGE_UQ 0x19 | |
127 | /* Not-greater-than (unordered, non-signaling) */ | |
128 | #define _CMP_NGT_UQ 0x1a | |
129 | /* False (ordered, signaling) */ | |
130 | #define _CMP_FALSE_OS 0x1b | |
131 | /* Not-equal (ordered, signaling) */ | |
132 | #define _CMP_NEQ_OS 0x1c | |
133 | /* Greater-than-or-equal (ordered, non-signaling) */ | |
134 | #define _CMP_GE_OQ 0x1d | |
135 | /* Greater-than (ordered, non-signaling) */ | |
136 | #define _CMP_GT_OQ 0x1e | |
137 | /* True (unordered, signaling) */ | |
138 | #define _CMP_TRUE_US 0x1f | |
139 | ||
140 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
141 | _mm256_add_pd (__m256d __A, __m256d __B) | |
142 | { | |
2069d6fc | 143 | return (__m256d) ((__v4df)__A + (__v4df)__B); |
95879c72 L |
144 | } |
145 | ||
146 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
147 | _mm256_add_ps (__m256 __A, __m256 __B) | |
148 | { | |
2069d6fc | 149 | return (__m256) ((__v8sf)__A + (__v8sf)__B); |
95879c72 L |
150 | } |
151 | ||
152 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
153 | _mm256_addsub_pd (__m256d __A, __m256d __B) | |
154 | { | |
155 | return (__m256d) __builtin_ia32_addsubpd256 ((__v4df)__A, (__v4df)__B); | |
156 | } | |
157 | ||
158 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
159 | _mm256_addsub_ps (__m256 __A, __m256 __B) | |
160 | { | |
161 | return (__m256) __builtin_ia32_addsubps256 ((__v8sf)__A, (__v8sf)__B); | |
162 | } | |
163 | ||
164 | ||
165 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
166 | _mm256_and_pd (__m256d __A, __m256d __B) | |
167 | { | |
168 | return (__m256d) __builtin_ia32_andpd256 ((__v4df)__A, (__v4df)__B); | |
169 | } | |
170 | ||
171 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
172 | _mm256_and_ps (__m256 __A, __m256 __B) | |
173 | { | |
174 | return (__m256) __builtin_ia32_andps256 ((__v8sf)__A, (__v8sf)__B); | |
175 | } | |
176 | ||
177 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
178 | _mm256_andnot_pd (__m256d __A, __m256d __B) | |
179 | { | |
180 | return (__m256d) __builtin_ia32_andnpd256 ((__v4df)__A, (__v4df)__B); | |
181 | } | |
182 | ||
183 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
184 | _mm256_andnot_ps (__m256 __A, __m256 __B) | |
185 | { | |
186 | return (__m256) __builtin_ia32_andnps256 ((__v8sf)__A, (__v8sf)__B); | |
187 | } | |
188 | ||
189 | /* Double/single precision floating point blend instructions - select | |
190 | data from 2 sources using constant/variable mask. */ | |
191 | ||
192 | #ifdef __OPTIMIZE__ | |
193 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
194 | _mm256_blend_pd (__m256d __X, __m256d __Y, const int __M) | |
195 | { | |
196 | return (__m256d) __builtin_ia32_blendpd256 ((__v4df)__X, | |
197 | (__v4df)__Y, | |
198 | __M); | |
199 | } | |
200 | ||
201 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
202 | _mm256_blend_ps (__m256 __X, __m256 __Y, const int __M) | |
203 | { | |
204 | return (__m256) __builtin_ia32_blendps256 ((__v8sf)__X, | |
205 | (__v8sf)__Y, | |
206 | __M); | |
207 | } | |
208 | #else | |
209 | #define _mm256_blend_pd(X, Y, M) \ | |
210 | ((__m256d) __builtin_ia32_blendpd256 ((__v4df)(__m256d)(X), \ | |
211 | (__v4df)(__m256d)(Y), (int)(M))) | |
212 | ||
213 | #define _mm256_blend_ps(X, Y, M) \ | |
214 | ((__m256) __builtin_ia32_blendps256 ((__v8sf)(__m256)(X), \ | |
215 | (__v8sf)(__m256)(Y), (int)(M))) | |
216 | #endif | |
217 | ||
218 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
219 | _mm256_blendv_pd (__m256d __X, __m256d __Y, __m256d __M) | |
220 | { | |
221 | return (__m256d) __builtin_ia32_blendvpd256 ((__v4df)__X, | |
222 | (__v4df)__Y, | |
223 | (__v4df)__M); | |
224 | } | |
225 | ||
226 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
227 | _mm256_blendv_ps (__m256 __X, __m256 __Y, __m256 __M) | |
228 | { | |
229 | return (__m256) __builtin_ia32_blendvps256 ((__v8sf)__X, | |
230 | (__v8sf)__Y, | |
231 | (__v8sf)__M); | |
232 | } | |
233 | ||
234 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
235 | _mm256_div_pd (__m256d __A, __m256d __B) | |
236 | { | |
2069d6fc | 237 | return (__m256d) ((__v4df)__A / (__v4df)__B); |
95879c72 L |
238 | } |
239 | ||
240 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
241 | _mm256_div_ps (__m256 __A, __m256 __B) | |
242 | { | |
2069d6fc | 243 | return (__m256) ((__v8sf)__A / (__v8sf)__B); |
95879c72 L |
244 | } |
245 | ||
246 | /* Dot product instructions with mask-defined summing and zeroing parts | |
247 | of result. */ | |
248 | ||
249 | #ifdef __OPTIMIZE__ | |
250 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
251 | _mm256_dp_ps (__m256 __X, __m256 __Y, const int __M) | |
252 | { | |
253 | return (__m256) __builtin_ia32_dpps256 ((__v8sf)__X, | |
254 | (__v8sf)__Y, | |
255 | __M); | |
256 | } | |
257 | #else | |
258 | #define _mm256_dp_ps(X, Y, M) \ | |
259 | ((__m256) __builtin_ia32_dpps256 ((__v8sf)(__m256)(X), \ | |
260 | (__v8sf)(__m256)(Y), (int)(M))) | |
261 | #endif | |
262 | ||
263 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
264 | _mm256_hadd_pd (__m256d __X, __m256d __Y) | |
265 | { | |
266 | return (__m256d) __builtin_ia32_haddpd256 ((__v4df)__X, (__v4df)__Y); | |
267 | } | |
268 | ||
269 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
270 | _mm256_hadd_ps (__m256 __X, __m256 __Y) | |
271 | { | |
272 | return (__m256) __builtin_ia32_haddps256 ((__v8sf)__X, (__v8sf)__Y); | |
273 | } | |
274 | ||
275 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
276 | _mm256_hsub_pd (__m256d __X, __m256d __Y) | |
277 | { | |
278 | return (__m256d) __builtin_ia32_hsubpd256 ((__v4df)__X, (__v4df)__Y); | |
279 | } | |
280 | ||
281 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
282 | _mm256_hsub_ps (__m256 __X, __m256 __Y) | |
283 | { | |
284 | return (__m256) __builtin_ia32_hsubps256 ((__v8sf)__X, (__v8sf)__Y); | |
285 | } | |
286 | ||
287 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
288 | _mm256_max_pd (__m256d __A, __m256d __B) | |
289 | { | |
290 | return (__m256d) __builtin_ia32_maxpd256 ((__v4df)__A, (__v4df)__B); | |
291 | } | |
292 | ||
293 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
294 | _mm256_max_ps (__m256 __A, __m256 __B) | |
295 | { | |
296 | return (__m256) __builtin_ia32_maxps256 ((__v8sf)__A, (__v8sf)__B); | |
297 | } | |
298 | ||
299 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
300 | _mm256_min_pd (__m256d __A, __m256d __B) | |
301 | { | |
302 | return (__m256d) __builtin_ia32_minpd256 ((__v4df)__A, (__v4df)__B); | |
303 | } | |
304 | ||
305 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
306 | _mm256_min_ps (__m256 __A, __m256 __B) | |
307 | { | |
308 | return (__m256) __builtin_ia32_minps256 ((__v8sf)__A, (__v8sf)__B); | |
309 | } | |
310 | ||
311 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
312 | _mm256_mul_pd (__m256d __A, __m256d __B) | |
313 | { | |
2069d6fc | 314 | return (__m256d) ((__v4df)__A * (__v4df)__B); |
95879c72 L |
315 | } |
316 | ||
317 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
318 | _mm256_mul_ps (__m256 __A, __m256 __B) | |
319 | { | |
2069d6fc | 320 | return (__m256) ((__v8sf)__A * (__v8sf)__B); |
95879c72 L |
321 | } |
322 | ||
323 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
324 | _mm256_or_pd (__m256d __A, __m256d __B) | |
325 | { | |
326 | return (__m256d) __builtin_ia32_orpd256 ((__v4df)__A, (__v4df)__B); | |
327 | } | |
328 | ||
329 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
330 | _mm256_or_ps (__m256 __A, __m256 __B) | |
331 | { | |
332 | return (__m256) __builtin_ia32_orps256 ((__v8sf)__A, (__v8sf)__B); | |
333 | } | |
334 | ||
335 | #ifdef __OPTIMIZE__ | |
336 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
337 | _mm256_shuffle_pd (__m256d __A, __m256d __B, const int __mask) | |
338 | { | |
339 | return (__m256d) __builtin_ia32_shufpd256 ((__v4df)__A, (__v4df)__B, | |
340 | __mask); | |
341 | } | |
342 | ||
343 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
344 | _mm256_shuffle_ps (__m256 __A, __m256 __B, const int __mask) | |
345 | { | |
346 | return (__m256) __builtin_ia32_shufps256 ((__v8sf)__A, (__v8sf)__B, | |
347 | __mask); | |
348 | } | |
349 | #else | |
350 | #define _mm256_shuffle_pd(A, B, N) \ | |
351 | ((__m256d)__builtin_ia32_shufpd256 ((__v4df)(__m256d)(A), \ | |
352 | (__v4df)(__m256d)(B), (int)(N))) | |
353 | ||
354 | #define _mm256_shuffle_ps(A, B, N) \ | |
355 | ((__m256) __builtin_ia32_shufps256 ((__v8sf)(__m256)(A), \ | |
356 | (__v8sf)(__m256)(B), (int)(N))) | |
357 | #endif | |
358 | ||
359 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
360 | _mm256_sub_pd (__m256d __A, __m256d __B) | |
361 | { | |
2069d6fc | 362 | return (__m256d) ((__v4df)__A - (__v4df)__B); |
95879c72 L |
363 | } |
364 | ||
365 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
366 | _mm256_sub_ps (__m256 __A, __m256 __B) | |
367 | { | |
2069d6fc | 368 | return (__m256) ((__v8sf)__A - (__v8sf)__B); |
95879c72 L |
369 | } |
370 | ||
371 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
372 | _mm256_xor_pd (__m256d __A, __m256d __B) | |
373 | { | |
374 | return (__m256d) __builtin_ia32_xorpd256 ((__v4df)__A, (__v4df)__B); | |
375 | } | |
376 | ||
377 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
378 | _mm256_xor_ps (__m256 __A, __m256 __B) | |
379 | { | |
380 | return (__m256) __builtin_ia32_xorps256 ((__v8sf)__A, (__v8sf)__B); | |
381 | } | |
382 | ||
383 | #ifdef __OPTIMIZE__ | |
384 | extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
385 | _mm_cmp_pd (__m128d __X, __m128d __Y, const int __P) | |
386 | { | |
387 | return (__m128d) __builtin_ia32_cmppd ((__v2df)__X, (__v2df)__Y, __P); | |
388 | } | |
389 | ||
390 | extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
391 | _mm_cmp_ps (__m128 __X, __m128 __Y, const int __P) | |
392 | { | |
393 | return (__m128) __builtin_ia32_cmpps ((__v4sf)__X, (__v4sf)__Y, __P); | |
394 | } | |
395 | ||
396 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
397 | _mm256_cmp_pd (__m256d __X, __m256d __Y, const int __P) | |
398 | { | |
399 | return (__m256d) __builtin_ia32_cmppd256 ((__v4df)__X, (__v4df)__Y, | |
400 | __P); | |
401 | } | |
402 | ||
403 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
404 | _mm256_cmp_ps (__m256 __X, __m256 __Y, const int __P) | |
405 | { | |
406 | return (__m256) __builtin_ia32_cmpps256 ((__v8sf)__X, (__v8sf)__Y, | |
407 | __P); | |
408 | } | |
409 | ||
410 | extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
411 | _mm_cmp_sd (__m128d __X, __m128d __Y, const int __P) | |
412 | { | |
413 | return (__m128d) __builtin_ia32_cmpsd ((__v2df)__X, (__v2df)__Y, __P); | |
414 | } | |
415 | ||
416 | extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
417 | _mm_cmp_ss (__m128 __X, __m128 __Y, const int __P) | |
418 | { | |
419 | return (__m128) __builtin_ia32_cmpss ((__v4sf)__X, (__v4sf)__Y, __P); | |
420 | } | |
421 | #else | |
422 | #define _mm_cmp_pd(X, Y, P) \ | |
423 | ((__m128d) __builtin_ia32_cmppd ((__v2df)(__m128d)(X), \ | |
424 | (__v2df)(__m128d)(Y), (int)(P))) | |
425 | ||
426 | #define _mm_cmp_ps(X, Y, P) \ | |
427 | ((__m128) __builtin_ia32_cmpps ((__v4sf)(__m128)(X), \ | |
428 | (__v4sf)(__m128)(Y), (int)(P))) | |
429 | ||
430 | #define _mm256_cmp_pd(X, Y, P) \ | |
431 | ((__m256d) __builtin_ia32_cmppd256 ((__v4df)(__m256d)(X), \ | |
432 | (__v4df)(__m256d)(Y), (int)(P))) | |
433 | ||
434 | #define _mm256_cmp_ps(X, Y, P) \ | |
435 | ((__m256) __builtin_ia32_cmpps256 ((__v8sf)(__m256)(X), \ | |
436 | (__v8sf)(__m256)(Y), (int)(P))) | |
437 | ||
438 | #define _mm_cmp_sd(X, Y, P) \ | |
439 | ((__m128d) __builtin_ia32_cmpsd ((__v2df)(__m128d)(X), \ | |
440 | (__v2df)(__m128d)(Y), (int)(P))) | |
441 | ||
442 | #define _mm_cmp_ss(X, Y, P) \ | |
443 | ((__m128) __builtin_ia32_cmpss ((__v4sf)(__m128)(X), \ | |
444 | (__v4sf)(__m128)(Y), (int)(P))) | |
445 | #endif | |
446 | ||
93103603 SP |
447 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
448 | _mm256_cvtsi256_si32 (__m256i __A) | |
449 | { | |
450 | __v8si __B = (__v8si) __A; | |
451 | return __B[0]; | |
452 | } | |
453 | ||
95879c72 L |
454 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
455 | _mm256_cvtepi32_pd (__m128i __A) | |
456 | { | |
457 | return (__m256d)__builtin_ia32_cvtdq2pd256 ((__v4si) __A); | |
458 | } | |
459 | ||
460 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
461 | _mm256_cvtepi32_ps (__m256i __A) | |
462 | { | |
463 | return (__m256)__builtin_ia32_cvtdq2ps256 ((__v8si) __A); | |
464 | } | |
465 | ||
466 | extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
467 | _mm256_cvtpd_ps (__m256d __A) | |
468 | { | |
469 | return (__m128)__builtin_ia32_cvtpd2ps256 ((__v4df) __A); | |
470 | } | |
471 | ||
472 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
473 | _mm256_cvtps_epi32 (__m256 __A) | |
474 | { | |
475 | return (__m256i)__builtin_ia32_cvtps2dq256 ((__v8sf) __A); | |
476 | } | |
477 | ||
478 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
479 | _mm256_cvtps_pd (__m128 __A) | |
480 | { | |
481 | return (__m256d)__builtin_ia32_cvtps2pd256 ((__v4sf) __A); | |
482 | } | |
483 | ||
484 | extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
485 | _mm256_cvttpd_epi32 (__m256d __A) | |
486 | { | |
487 | return (__m128i)__builtin_ia32_cvttpd2dq256 ((__v4df) __A); | |
488 | } | |
489 | ||
490 | extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
491 | _mm256_cvtpd_epi32 (__m256d __A) | |
492 | { | |
493 | return (__m128i)__builtin_ia32_cvtpd2dq256 ((__v4df) __A); | |
494 | } | |
495 | ||
496 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
497 | _mm256_cvttps_epi32 (__m256 __A) | |
498 | { | |
499 | return (__m256i)__builtin_ia32_cvttps2dq256 ((__v8sf) __A); | |
500 | } | |
501 | ||
dcb2c527 JJ |
502 | extern __inline double |
503 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) | |
504 | _mm256_cvtsd_f64 (__m256d __A) | |
505 | { | |
506 | return __A[0]; | |
507 | } | |
508 | ||
509 | extern __inline float | |
510 | __attribute__ ((__gnu_inline__, __always_inline__, __artificial__)) | |
511 | _mm256_cvtss_f32 (__m256 __A) | |
512 | { | |
513 | return __A[0]; | |
514 | } | |
515 | ||
95879c72 L |
516 | #ifdef __OPTIMIZE__ |
517 | extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
518 | _mm256_extractf128_pd (__m256d __X, const int __N) | |
519 | { | |
520 | return (__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)__X, __N); | |
521 | } | |
522 | ||
523 | extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
524 | _mm256_extractf128_ps (__m256 __X, const int __N) | |
525 | { | |
526 | return (__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)__X, __N); | |
527 | } | |
528 | ||
529 | extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
530 | _mm256_extractf128_si256 (__m256i __X, const int __N) | |
531 | { | |
532 | return (__m128i) __builtin_ia32_vextractf128_si256 ((__v8si)__X, __N); | |
533 | } | |
534 | ||
535 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
536 | _mm256_extract_epi32 (__m256i __X, int const __N) | |
537 | { | |
538 | __m128i __Y = _mm256_extractf128_si256 (__X, __N >> 2); | |
539 | return _mm_extract_epi32 (__Y, __N % 4); | |
540 | } | |
541 | ||
542 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
543 | _mm256_extract_epi16 (__m256i __X, int const __N) | |
544 | { | |
545 | __m128i __Y = _mm256_extractf128_si256 (__X, __N >> 3); | |
546 | return _mm_extract_epi16 (__Y, __N % 8); | |
547 | } | |
548 | ||
549 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
550 | _mm256_extract_epi8 (__m256i __X, int const __N) | |
551 | { | |
552 | __m128i __Y = _mm256_extractf128_si256 (__X, __N >> 4); | |
553 | return _mm_extract_epi8 (__Y, __N % 16); | |
554 | } | |
555 | ||
556 | #ifdef __x86_64__ | |
557 | extern __inline long long __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
558 | _mm256_extract_epi64 (__m256i __X, const int __N) | |
559 | { | |
560 | __m128i __Y = _mm256_extractf128_si256 (__X, __N >> 1); | |
561 | return _mm_extract_epi64 (__Y, __N % 2); | |
562 | } | |
563 | #endif | |
564 | #else | |
565 | #define _mm256_extractf128_pd(X, N) \ | |
566 | ((__m128d) __builtin_ia32_vextractf128_pd256 ((__v4df)(__m256d)(X), \ | |
567 | (int)(N))) | |
568 | ||
569 | #define _mm256_extractf128_ps(X, N) \ | |
570 | ((__m128) __builtin_ia32_vextractf128_ps256 ((__v8sf)(__m256)(X), \ | |
571 | (int)(N))) | |
572 | ||
573 | #define _mm256_extractf128_si256(X, N) \ | |
574 | ((__m128i) __builtin_ia32_vextractf128_si256 ((__v8si)(__m256i)(X), \ | |
575 | (int)(N))) | |
576 | ||
577 | #define _mm256_extract_epi32(X, N) \ | |
578 | (__extension__ \ | |
579 | ({ \ | |
580 | __m128i __Y = _mm256_extractf128_si256 ((X), (N) >> 2); \ | |
581 | _mm_extract_epi32 (__Y, (N) % 4); \ | |
582 | })) | |
583 | ||
584 | #define _mm256_extract_epi16(X, N) \ | |
585 | (__extension__ \ | |
586 | ({ \ | |
587 | __m128i __Y = _mm256_extractf128_si256 ((X), (N) >> 3); \ | |
588 | _mm_extract_epi16 (__Y, (N) % 8); \ | |
589 | })) | |
590 | ||
591 | #define _mm256_extract_epi8(X, N) \ | |
592 | (__extension__ \ | |
593 | ({ \ | |
594 | __m128i __Y = _mm256_extractf128_si256 ((X), (N) >> 4); \ | |
595 | _mm_extract_epi8 (__Y, (N) % 16); \ | |
596 | })) | |
597 | ||
598 | #ifdef __x86_64__ | |
599 | #define _mm256_extract_epi64(X, N) \ | |
600 | (__extension__ \ | |
601 | ({ \ | |
602 | __m128i __Y = _mm256_extractf128_si256 ((X), (N) >> 1); \ | |
603 | _mm_extract_epi64 (__Y, (N) % 2); \ | |
604 | })) | |
605 | #endif | |
606 | #endif | |
607 | ||
608 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
609 | _mm256_zeroall (void) | |
610 | { | |
611 | __builtin_ia32_vzeroall (); | |
612 | } | |
613 | ||
614 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
615 | _mm256_zeroupper (void) | |
616 | { | |
617 | __builtin_ia32_vzeroupper (); | |
618 | } | |
619 | ||
620 | extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
621 | _mm_permutevar_pd (__m128d __A, __m128i __C) | |
622 | { | |
623 | return (__m128d) __builtin_ia32_vpermilvarpd ((__v2df)__A, | |
624 | (__v2di)__C); | |
625 | } | |
626 | ||
627 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
628 | _mm256_permutevar_pd (__m256d __A, __m256i __C) | |
629 | { | |
630 | return (__m256d) __builtin_ia32_vpermilvarpd256 ((__v4df)__A, | |
631 | (__v4di)__C); | |
632 | } | |
633 | ||
634 | extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
635 | _mm_permutevar_ps (__m128 __A, __m128i __C) | |
636 | { | |
637 | return (__m128) __builtin_ia32_vpermilvarps ((__v4sf)__A, | |
638 | (__v4si)__C); | |
639 | } | |
640 | ||
641 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
642 | _mm256_permutevar_ps (__m256 __A, __m256i __C) | |
643 | { | |
644 | return (__m256) __builtin_ia32_vpermilvarps256 ((__v8sf)__A, | |
645 | (__v8si)__C); | |
646 | } | |
647 | ||
648 | #ifdef __OPTIMIZE__ | |
649 | extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
650 | _mm_permute_pd (__m128d __X, const int __C) | |
651 | { | |
652 | return (__m128d) __builtin_ia32_vpermilpd ((__v2df)__X, __C); | |
653 | } | |
654 | ||
655 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
656 | _mm256_permute_pd (__m256d __X, const int __C) | |
657 | { | |
658 | return (__m256d) __builtin_ia32_vpermilpd256 ((__v4df)__X, __C); | |
659 | } | |
660 | ||
661 | extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
662 | _mm_permute_ps (__m128 __X, const int __C) | |
663 | { | |
664 | return (__m128) __builtin_ia32_vpermilps ((__v4sf)__X, __C); | |
665 | } | |
666 | ||
667 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
668 | _mm256_permute_ps (__m256 __X, const int __C) | |
669 | { | |
670 | return (__m256) __builtin_ia32_vpermilps256 ((__v8sf)__X, __C); | |
671 | } | |
95879c72 L |
672 | #else |
673 | #define _mm_permute_pd(X, C) \ | |
674 | ((__m128d) __builtin_ia32_vpermilpd ((__v2df)(__m128d)(X), (int)(C))) | |
675 | ||
676 | #define _mm256_permute_pd(X, C) \ | |
677 | ((__m256d) __builtin_ia32_vpermilpd256 ((__v4df)(__m256d)(X), (int)(C))) | |
678 | ||
679 | #define _mm_permute_ps(X, C) \ | |
680 | ((__m128) __builtin_ia32_vpermilps ((__v4sf)(__m128)(X), (int)(C))) | |
681 | ||
682 | #define _mm256_permute_ps(X, C) \ | |
683 | ((__m256) __builtin_ia32_vpermilps256 ((__v8sf)(__m256)(X), (int)(C))) | |
95879c72 L |
684 | #endif |
685 | ||
686 | #ifdef __OPTIMIZE__ | |
687 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
688 | _mm256_permute2f128_pd (__m256d __X, __m256d __Y, const int __C) | |
689 | { | |
690 | return (__m256d) __builtin_ia32_vperm2f128_pd256 ((__v4df)__X, | |
691 | (__v4df)__Y, | |
692 | __C); | |
693 | } | |
694 | ||
695 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
696 | _mm256_permute2f128_ps (__m256 __X, __m256 __Y, const int __C) | |
697 | { | |
698 | return (__m256) __builtin_ia32_vperm2f128_ps256 ((__v8sf)__X, | |
699 | (__v8sf)__Y, | |
700 | __C); | |
701 | } | |
702 | ||
703 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
704 | _mm256_permute2f128_si256 (__m256i __X, __m256i __Y, const int __C) | |
705 | { | |
706 | return (__m256i) __builtin_ia32_vperm2f128_si256 ((__v8si)__X, | |
707 | (__v8si)__Y, | |
708 | __C); | |
709 | } | |
710 | #else | |
711 | #define _mm256_permute2f128_pd(X, Y, C) \ | |
712 | ((__m256d) __builtin_ia32_vperm2f128_pd256 ((__v4df)(__m256d)(X), \ | |
713 | (__v4df)(__m256d)(Y), \ | |
714 | (int)(C))) | |
715 | ||
716 | #define _mm256_permute2f128_ps(X, Y, C) \ | |
717 | ((__m256) __builtin_ia32_vperm2f128_ps256 ((__v8sf)(__m256)(X), \ | |
718 | (__v8sf)(__m256)(Y), \ | |
719 | (int)(C))) | |
720 | ||
721 | #define _mm256_permute2f128_si256(X, Y, C) \ | |
722 | ((__m256i) __builtin_ia32_vperm2f128_si256 ((__v8si)(__m256i)(X), \ | |
723 | (__v8si)(__m256i)(Y), \ | |
724 | (int)(C))) | |
725 | #endif | |
726 | ||
727 | extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
728 | _mm_broadcast_ss (float const *__X) | |
729 | { | |
730 | return (__m128) __builtin_ia32_vbroadcastss (__X); | |
731 | } | |
732 | ||
733 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
734 | _mm256_broadcast_sd (double const *__X) | |
735 | { | |
736 | return (__m256d) __builtin_ia32_vbroadcastsd256 (__X); | |
737 | } | |
738 | ||
739 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
740 | _mm256_broadcast_ss (float const *__X) | |
741 | { | |
742 | return (__m256) __builtin_ia32_vbroadcastss256 (__X); | |
743 | } | |
744 | ||
745 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
746 | _mm256_broadcast_pd (__m128d const *__X) | |
747 | { | |
748 | return (__m256d) __builtin_ia32_vbroadcastf128_pd256 (__X); | |
749 | } | |
750 | ||
751 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
752 | _mm256_broadcast_ps (__m128 const *__X) | |
753 | { | |
754 | return (__m256) __builtin_ia32_vbroadcastf128_ps256 (__X); | |
755 | } | |
756 | ||
757 | #ifdef __OPTIMIZE__ | |
758 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
759 | _mm256_insertf128_pd (__m256d __X, __m128d __Y, const int __O) | |
760 | { | |
761 | return (__m256d) __builtin_ia32_vinsertf128_pd256 ((__v4df)__X, | |
762 | (__v2df)__Y, | |
763 | __O); | |
764 | } | |
765 | ||
766 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
767 | _mm256_insertf128_ps (__m256 __X, __m128 __Y, const int __O) | |
768 | { | |
769 | return (__m256) __builtin_ia32_vinsertf128_ps256 ((__v8sf)__X, | |
770 | (__v4sf)__Y, | |
771 | __O); | |
772 | } | |
773 | ||
774 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
775 | _mm256_insertf128_si256 (__m256i __X, __m128i __Y, const int __O) | |
776 | { | |
777 | return (__m256i) __builtin_ia32_vinsertf128_si256 ((__v8si)__X, | |
778 | (__v4si)__Y, | |
779 | __O); | |
780 | } | |
781 | ||
782 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
783 | _mm256_insert_epi32 (__m256i __X, int __D, int const __N) | |
784 | { | |
785 | __m128i __Y = _mm256_extractf128_si256 (__X, __N >> 2); | |
23e0d930 | 786 | __Y = _mm_insert_epi32 (__Y, __D, __N % 4); |
95879c72 L |
787 | return _mm256_insertf128_si256 (__X, __Y, __N >> 2); |
788 | } | |
789 | ||
790 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
791 | _mm256_insert_epi16 (__m256i __X, int __D, int const __N) | |
792 | { | |
793 | __m128i __Y = _mm256_extractf128_si256 (__X, __N >> 3); | |
794 | __Y = _mm_insert_epi16 (__Y, __D, __N % 8); | |
795 | return _mm256_insertf128_si256 (__X, __Y, __N >> 3); | |
796 | } | |
797 | ||
798 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
799 | _mm256_insert_epi8 (__m256i __X, int __D, int const __N) | |
800 | { | |
801 | __m128i __Y = _mm256_extractf128_si256 (__X, __N >> 4); | |
802 | __Y = _mm_insert_epi8 (__Y, __D, __N % 16); | |
803 | return _mm256_insertf128_si256 (__X, __Y, __N >> 4); | |
804 | } | |
805 | ||
806 | #ifdef __x86_64__ | |
807 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
fcff2e9c | 808 | _mm256_insert_epi64 (__m256i __X, long long __D, int const __N) |
95879c72 L |
809 | { |
810 | __m128i __Y = _mm256_extractf128_si256 (__X, __N >> 1); | |
23e0d930 | 811 | __Y = _mm_insert_epi64 (__Y, __D, __N % 2); |
95879c72 L |
812 | return _mm256_insertf128_si256 (__X, __Y, __N >> 1); |
813 | } | |
814 | #endif | |
815 | #else | |
816 | #define _mm256_insertf128_pd(X, Y, O) \ | |
817 | ((__m256d) __builtin_ia32_vinsertf128_pd256 ((__v4df)(__m256d)(X), \ | |
818 | (__v2df)(__m128d)(Y), \ | |
819 | (int)(O))) | |
820 | ||
821 | #define _mm256_insertf128_ps(X, Y, O) \ | |
822 | ((__m256) __builtin_ia32_vinsertf128_ps256 ((__v8sf)(__m256)(X), \ | |
823 | (__v4sf)(__m128)(Y), \ | |
824 | (int)(O))) | |
825 | ||
826 | #define _mm256_insertf128_si256(X, Y, O) \ | |
827 | ((__m256i) __builtin_ia32_vinsertf128_si256 ((__v8si)(__m256i)(X), \ | |
828 | (__v4si)(__m128i)(Y), \ | |
829 | (int)(O))) | |
830 | ||
831 | #define _mm256_insert_epi32(X, D, N) \ | |
832 | (__extension__ \ | |
833 | ({ \ | |
834 | __m128i __Y = _mm256_extractf128_si256 ((X), (N) >> 2); \ | |
835 | __Y = _mm_insert_epi32 (__Y, (D), (N) % 4); \ | |
836 | _mm256_insertf128_si256 ((X), __Y, (N) >> 2); \ | |
837 | })) | |
838 | ||
839 | #define _mm256_insert_epi16(X, D, N) \ | |
840 | (__extension__ \ | |
841 | ({ \ | |
842 | __m128i __Y = _mm256_extractf128_si256 ((X), (N) >> 3); \ | |
843 | __Y = _mm_insert_epi16 (__Y, (D), (N) % 8); \ | |
844 | _mm256_insertf128_si256 ((X), __Y, (N) >> 3); \ | |
845 | })) | |
846 | ||
847 | #define _mm256_insert_epi8(X, D, N) \ | |
848 | (__extension__ \ | |
849 | ({ \ | |
850 | __m128i __Y = _mm256_extractf128_si256 ((X), (N) >> 4); \ | |
851 | __Y = _mm_insert_epi8 (__Y, (D), (N) % 16); \ | |
852 | _mm256_insertf128_si256 ((X), __Y, (N) >> 4); \ | |
853 | })) | |
854 | ||
855 | #ifdef __x86_64__ | |
856 | #define _mm256_insert_epi64(X, D, N) \ | |
857 | (__extension__ \ | |
858 | ({ \ | |
859 | __m128i __Y = _mm256_extractf128_si256 ((X), (N) >> 1); \ | |
860 | __Y = _mm_insert_epi64 (__Y, (D), (N) % 2); \ | |
861 | _mm256_insertf128_si256 ((X), __Y, (N) >> 1); \ | |
862 | })) | |
863 | #endif | |
864 | #endif | |
865 | ||
866 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
867 | _mm256_load_pd (double const *__P) | |
868 | { | |
869 | return *(__m256d *)__P; | |
870 | } | |
871 | ||
872 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
873 | _mm256_store_pd (double *__P, __m256d __A) | |
874 | { | |
875 | *(__m256d *)__P = __A; | |
876 | } | |
877 | ||
878 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
879 | _mm256_load_ps (float const *__P) | |
880 | { | |
881 | return *(__m256 *)__P; | |
882 | } | |
883 | ||
884 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
885 | _mm256_store_ps (float *__P, __m256 __A) | |
886 | { | |
887 | *(__m256 *)__P = __A; | |
888 | } | |
889 | ||
890 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
891 | _mm256_loadu_pd (double const *__P) | |
892 | { | |
c6b0037d | 893 | return *(__m256d_u *)__P; |
95879c72 L |
894 | } |
895 | ||
896 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
897 | _mm256_storeu_pd (double *__P, __m256d __A) | |
898 | { | |
c6b0037d | 899 | *(__m256d_u *)__P = __A; |
95879c72 L |
900 | } |
901 | ||
902 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
903 | _mm256_loadu_ps (float const *__P) | |
904 | { | |
c6b0037d | 905 | return *(__m256_u *)__P; |
95879c72 L |
906 | } |
907 | ||
908 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
909 | _mm256_storeu_ps (float *__P, __m256 __A) | |
910 | { | |
c6b0037d | 911 | *(__m256_u *)__P = __A; |
95879c72 L |
912 | } |
913 | ||
914 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
915 | _mm256_load_si256 (__m256i const *__P) | |
916 | { | |
917 | return *__P; | |
918 | } | |
919 | ||
920 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
921 | _mm256_store_si256 (__m256i *__P, __m256i __A) | |
922 | { | |
923 | *__P = __A; | |
924 | } | |
925 | ||
926 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
c6b0037d | 927 | _mm256_loadu_si256 (__m256i_u const *__P) |
95879c72 | 928 | { |
c6b0037d | 929 | return *__P; |
95879c72 L |
930 | } |
931 | ||
932 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
c6b0037d | 933 | _mm256_storeu_si256 (__m256i_u *__P, __m256i __A) |
95879c72 | 934 | { |
c6b0037d | 935 | *__P = __A; |
95879c72 L |
936 | } |
937 | ||
938 | extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
98c6d93c | 939 | _mm_maskload_pd (double const *__P, __m128i __M) |
95879c72 L |
940 | { |
941 | return (__m128d) __builtin_ia32_maskloadpd ((const __v2df *)__P, | |
98c6d93c | 942 | (__v2di)__M); |
95879c72 L |
943 | } |
944 | ||
945 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
98c6d93c | 946 | _mm_maskstore_pd (double *__P, __m128i __M, __m128d __A) |
95879c72 | 947 | { |
98c6d93c | 948 | __builtin_ia32_maskstorepd ((__v2df *)__P, (__v2di)__M, (__v2df)__A); |
95879c72 L |
949 | } |
950 | ||
951 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
98c6d93c | 952 | _mm256_maskload_pd (double const *__P, __m256i __M) |
95879c72 L |
953 | { |
954 | return (__m256d) __builtin_ia32_maskloadpd256 ((const __v4df *)__P, | |
98c6d93c | 955 | (__v4di)__M); |
95879c72 L |
956 | } |
957 | ||
958 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
98c6d93c | 959 | _mm256_maskstore_pd (double *__P, __m256i __M, __m256d __A) |
95879c72 | 960 | { |
98c6d93c | 961 | __builtin_ia32_maskstorepd256 ((__v4df *)__P, (__v4di)__M, (__v4df)__A); |
95879c72 L |
962 | } |
963 | ||
964 | extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
98c6d93c | 965 | _mm_maskload_ps (float const *__P, __m128i __M) |
95879c72 L |
966 | { |
967 | return (__m128) __builtin_ia32_maskloadps ((const __v4sf *)__P, | |
98c6d93c | 968 | (__v4si)__M); |
95879c72 L |
969 | } |
970 | ||
971 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
98c6d93c | 972 | _mm_maskstore_ps (float *__P, __m128i __M, __m128 __A) |
95879c72 | 973 | { |
98c6d93c | 974 | __builtin_ia32_maskstoreps ((__v4sf *)__P, (__v4si)__M, (__v4sf)__A); |
95879c72 L |
975 | } |
976 | ||
977 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
98c6d93c | 978 | _mm256_maskload_ps (float const *__P, __m256i __M) |
95879c72 L |
979 | { |
980 | return (__m256) __builtin_ia32_maskloadps256 ((const __v8sf *)__P, | |
98c6d93c | 981 | (__v8si)__M); |
95879c72 L |
982 | } |
983 | ||
984 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
98c6d93c | 985 | _mm256_maskstore_ps (float *__P, __m256i __M, __m256 __A) |
95879c72 | 986 | { |
98c6d93c | 987 | __builtin_ia32_maskstoreps256 ((__v8sf *)__P, (__v8si)__M, (__v8sf)__A); |
95879c72 L |
988 | } |
989 | ||
990 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
991 | _mm256_movehdup_ps (__m256 __X) | |
992 | { | |
993 | return (__m256) __builtin_ia32_movshdup256 ((__v8sf)__X); | |
994 | } | |
995 | ||
996 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
997 | _mm256_moveldup_ps (__m256 __X) | |
998 | { | |
999 | return (__m256) __builtin_ia32_movsldup256 ((__v8sf)__X); | |
1000 | } | |
1001 | ||
1002 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1003 | _mm256_movedup_pd (__m256d __X) | |
1004 | { | |
1005 | return (__m256d) __builtin_ia32_movddup256 ((__v4df)__X); | |
1006 | } | |
1007 | ||
1008 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1009 | _mm256_lddqu_si256 (__m256i const *__P) | |
1010 | { | |
1011 | return (__m256i) __builtin_ia32_lddqu256 ((char const *)__P); | |
1012 | } | |
1013 | ||
65b82caa L |
1014 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
1015 | _mm256_stream_si256 (__m256i *__A, __m256i __B) | |
1016 | { | |
1017 | __builtin_ia32_movntdq256 ((__v4di *)__A, (__v4di)__B); | |
1018 | } | |
1019 | ||
1020 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1021 | _mm256_stream_pd (double *__A, __m256d __B) | |
1022 | { | |
1023 | __builtin_ia32_movntpd256 (__A, (__v4df)__B); | |
1024 | } | |
1025 | ||
1026 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1027 | _mm256_stream_ps (float *__P, __m256 __A) | |
1028 | { | |
1029 | __builtin_ia32_movntps256 (__P, (__v8sf)__A); | |
1030 | } | |
1031 | ||
95879c72 L |
1032 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
1033 | _mm256_rcp_ps (__m256 __A) | |
1034 | { | |
1035 | return (__m256) __builtin_ia32_rcpps256 ((__v8sf)__A); | |
1036 | } | |
1037 | ||
1038 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1039 | _mm256_rsqrt_ps (__m256 __A) | |
1040 | { | |
1041 | return (__m256) __builtin_ia32_rsqrtps256 ((__v8sf)__A); | |
1042 | } | |
1043 | ||
1044 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1045 | _mm256_sqrt_pd (__m256d __A) | |
1046 | { | |
1047 | return (__m256d) __builtin_ia32_sqrtpd256 ((__v4df)__A); | |
1048 | } | |
1049 | ||
1050 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1051 | _mm256_sqrt_ps (__m256 __A) | |
1052 | { | |
1053 | return (__m256) __builtin_ia32_sqrtps256 ((__v8sf)__A); | |
1054 | } | |
1055 | ||
1056 | #ifdef __OPTIMIZE__ | |
1057 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1058 | _mm256_round_pd (__m256d __V, const int __M) | |
1059 | { | |
1060 | return (__m256d) __builtin_ia32_roundpd256 ((__v4df)__V, __M); | |
1061 | } | |
1062 | ||
1063 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1064 | _mm256_round_ps (__m256 __V, const int __M) | |
1065 | { | |
1066 | return (__m256) __builtin_ia32_roundps256 ((__v8sf)__V, __M); | |
1067 | } | |
1068 | #else | |
1069 | #define _mm256_round_pd(V, M) \ | |
1070 | ((__m256d) __builtin_ia32_roundpd256 ((__v4df)(__m256d)(V), (int)(M))) | |
1071 | ||
1072 | #define _mm256_round_ps(V, M) \ | |
1073 | ((__m256) __builtin_ia32_roundps256 ((__v8sf)(__m256)(V), (int)(M))) | |
1074 | #endif | |
1075 | ||
1076 | #define _mm256_ceil_pd(V) _mm256_round_pd ((V), _MM_FROUND_CEIL) | |
1077 | #define _mm256_floor_pd(V) _mm256_round_pd ((V), _MM_FROUND_FLOOR) | |
1078 | #define _mm256_ceil_ps(V) _mm256_round_ps ((V), _MM_FROUND_CEIL) | |
1079 | #define _mm256_floor_ps(V) _mm256_round_ps ((V), _MM_FROUND_FLOOR) | |
1080 | ||
1081 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1082 | _mm256_unpackhi_pd (__m256d __A, __m256d __B) | |
1083 | { | |
1084 | return (__m256d) __builtin_ia32_unpckhpd256 ((__v4df)__A, (__v4df)__B); | |
1085 | } | |
1086 | ||
1087 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1088 | _mm256_unpacklo_pd (__m256d __A, __m256d __B) | |
1089 | { | |
1090 | return (__m256d) __builtin_ia32_unpcklpd256 ((__v4df)__A, (__v4df)__B); | |
1091 | } | |
1092 | ||
1093 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1094 | _mm256_unpackhi_ps (__m256 __A, __m256 __B) | |
1095 | { | |
1096 | return (__m256) __builtin_ia32_unpckhps256 ((__v8sf)__A, (__v8sf)__B); | |
1097 | } | |
1098 | ||
1099 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1100 | _mm256_unpacklo_ps (__m256 __A, __m256 __B) | |
1101 | { | |
1102 | return (__m256) __builtin_ia32_unpcklps256 ((__v8sf)__A, (__v8sf)__B); | |
1103 | } | |
1104 | ||
1105 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1106 | _mm_testz_pd (__m128d __M, __m128d __V) | |
1107 | { | |
1108 | return __builtin_ia32_vtestzpd ((__v2df)__M, (__v2df)__V); | |
1109 | } | |
1110 | ||
1111 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1112 | _mm_testc_pd (__m128d __M, __m128d __V) | |
1113 | { | |
1114 | return __builtin_ia32_vtestcpd ((__v2df)__M, (__v2df)__V); | |
1115 | } | |
1116 | ||
1117 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1118 | _mm_testnzc_pd (__m128d __M, __m128d __V) | |
1119 | { | |
1120 | return __builtin_ia32_vtestnzcpd ((__v2df)__M, (__v2df)__V); | |
1121 | } | |
1122 | ||
1123 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1124 | _mm_testz_ps (__m128 __M, __m128 __V) | |
1125 | { | |
1126 | return __builtin_ia32_vtestzps ((__v4sf)__M, (__v4sf)__V); | |
1127 | } | |
1128 | ||
1129 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1130 | _mm_testc_ps (__m128 __M, __m128 __V) | |
1131 | { | |
1132 | return __builtin_ia32_vtestcps ((__v4sf)__M, (__v4sf)__V); | |
1133 | } | |
1134 | ||
1135 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1136 | _mm_testnzc_ps (__m128 __M, __m128 __V) | |
1137 | { | |
1138 | return __builtin_ia32_vtestnzcps ((__v4sf)__M, (__v4sf)__V); | |
1139 | } | |
1140 | ||
1141 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1142 | _mm256_testz_pd (__m256d __M, __m256d __V) | |
1143 | { | |
1144 | return __builtin_ia32_vtestzpd256 ((__v4df)__M, (__v4df)__V); | |
1145 | } | |
1146 | ||
1147 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1148 | _mm256_testc_pd (__m256d __M, __m256d __V) | |
1149 | { | |
1150 | return __builtin_ia32_vtestcpd256 ((__v4df)__M, (__v4df)__V); | |
1151 | } | |
1152 | ||
1153 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1154 | _mm256_testnzc_pd (__m256d __M, __m256d __V) | |
1155 | { | |
1156 | return __builtin_ia32_vtestnzcpd256 ((__v4df)__M, (__v4df)__V); | |
1157 | } | |
1158 | ||
1159 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1160 | _mm256_testz_ps (__m256 __M, __m256 __V) | |
1161 | { | |
1162 | return __builtin_ia32_vtestzps256 ((__v8sf)__M, (__v8sf)__V); | |
1163 | } | |
1164 | ||
1165 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1166 | _mm256_testc_ps (__m256 __M, __m256 __V) | |
1167 | { | |
1168 | return __builtin_ia32_vtestcps256 ((__v8sf)__M, (__v8sf)__V); | |
1169 | } | |
1170 | ||
1171 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1172 | _mm256_testnzc_ps (__m256 __M, __m256 __V) | |
1173 | { | |
1174 | return __builtin_ia32_vtestnzcps256 ((__v8sf)__M, (__v8sf)__V); | |
1175 | } | |
1176 | ||
1177 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1178 | _mm256_testz_si256 (__m256i __M, __m256i __V) | |
1179 | { | |
1180 | return __builtin_ia32_ptestz256 ((__v4di)__M, (__v4di)__V); | |
1181 | } | |
1182 | ||
1183 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1184 | _mm256_testc_si256 (__m256i __M, __m256i __V) | |
1185 | { | |
1186 | return __builtin_ia32_ptestc256 ((__v4di)__M, (__v4di)__V); | |
1187 | } | |
1188 | ||
1189 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1190 | _mm256_testnzc_si256 (__m256i __M, __m256i __V) | |
1191 | { | |
1192 | return __builtin_ia32_ptestnzc256 ((__v4di)__M, (__v4di)__V); | |
1193 | } | |
1194 | ||
1195 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1196 | _mm256_movemask_pd (__m256d __A) | |
1197 | { | |
1198 | return __builtin_ia32_movmskpd256 ((__v4df)__A); | |
1199 | } | |
1200 | ||
1201 | extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1202 | _mm256_movemask_ps (__m256 __A) | |
1203 | { | |
1204 | return __builtin_ia32_movmskps256 ((__v8sf)__A); | |
1205 | } | |
1206 | ||
0b192937 UD |
1207 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
1208 | _mm256_undefined_pd (void) | |
1209 | { | |
6b0907b4 JJ |
1210 | #pragma GCC diagnostic push |
1211 | #pragma GCC diagnostic ignored "-Winit-self" | |
0b192937 | 1212 | __m256d __Y = __Y; |
6b0907b4 | 1213 | #pragma GCC diagnostic pop |
0b192937 UD |
1214 | return __Y; |
1215 | } | |
1216 | ||
1217 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1218 | _mm256_undefined_ps (void) | |
1219 | { | |
6b0907b4 JJ |
1220 | #pragma GCC diagnostic push |
1221 | #pragma GCC diagnostic ignored "-Winit-self" | |
0b192937 | 1222 | __m256 __Y = __Y; |
6b0907b4 | 1223 | #pragma GCC diagnostic pop |
0b192937 UD |
1224 | return __Y; |
1225 | } | |
1226 | ||
1227 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1228 | _mm256_undefined_si256 (void) | |
1229 | { | |
6b0907b4 JJ |
1230 | #pragma GCC diagnostic push |
1231 | #pragma GCC diagnostic ignored "-Winit-self" | |
0b192937 | 1232 | __m256i __Y = __Y; |
6b0907b4 | 1233 | #pragma GCC diagnostic pop |
0b192937 UD |
1234 | return __Y; |
1235 | } | |
1236 | ||
95879c72 L |
1237 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
1238 | _mm256_setzero_pd (void) | |
1239 | { | |
1240 | return __extension__ (__m256d){ 0.0, 0.0, 0.0, 0.0 }; | |
1241 | } | |
1242 | ||
1243 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1244 | _mm256_setzero_ps (void) | |
1245 | { | |
1246 | return __extension__ (__m256){ 0.0, 0.0, 0.0, 0.0, | |
1247 | 0.0, 0.0, 0.0, 0.0 }; | |
1248 | } | |
1249 | ||
1250 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1251 | _mm256_setzero_si256 (void) | |
1252 | { | |
1253 | return __extension__ (__m256i)(__v4di){ 0, 0, 0, 0 }; | |
1254 | } | |
1255 | ||
1256 | /* Create the vector [A B C D]. */ | |
1257 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1258 | _mm256_set_pd (double __A, double __B, double __C, double __D) | |
1259 | { | |
1260 | return __extension__ (__m256d){ __D, __C, __B, __A }; | |
1261 | } | |
1262 | ||
1263 | /* Create the vector [A B C D E F G H]. */ | |
1264 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1265 | _mm256_set_ps (float __A, float __B, float __C, float __D, | |
1266 | float __E, float __F, float __G, float __H) | |
1267 | { | |
1268 | return __extension__ (__m256){ __H, __G, __F, __E, | |
1269 | __D, __C, __B, __A }; | |
1270 | } | |
1271 | ||
1272 | /* Create the vector [A B C D E F G H]. */ | |
1273 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1274 | _mm256_set_epi32 (int __A, int __B, int __C, int __D, | |
1275 | int __E, int __F, int __G, int __H) | |
1276 | { | |
1277 | return __extension__ (__m256i)(__v8si){ __H, __G, __F, __E, | |
1278 | __D, __C, __B, __A }; | |
1279 | } | |
1280 | ||
1281 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1282 | _mm256_set_epi16 (short __q15, short __q14, short __q13, short __q12, | |
1283 | short __q11, short __q10, short __q09, short __q08, | |
1284 | short __q07, short __q06, short __q05, short __q04, | |
1285 | short __q03, short __q02, short __q01, short __q00) | |
1286 | { | |
1287 | return __extension__ (__m256i)(__v16hi){ | |
1288 | __q00, __q01, __q02, __q03, __q04, __q05, __q06, __q07, | |
1289 | __q08, __q09, __q10, __q11, __q12, __q13, __q14, __q15 | |
1290 | }; | |
1291 | } | |
1292 | ||
1293 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1294 | _mm256_set_epi8 (char __q31, char __q30, char __q29, char __q28, | |
1295 | char __q27, char __q26, char __q25, char __q24, | |
1296 | char __q23, char __q22, char __q21, char __q20, | |
1297 | char __q19, char __q18, char __q17, char __q16, | |
1298 | char __q15, char __q14, char __q13, char __q12, | |
1299 | char __q11, char __q10, char __q09, char __q08, | |
1300 | char __q07, char __q06, char __q05, char __q04, | |
1301 | char __q03, char __q02, char __q01, char __q00) | |
1302 | { | |
1303 | return __extension__ (__m256i)(__v32qi){ | |
1304 | __q00, __q01, __q02, __q03, __q04, __q05, __q06, __q07, | |
1305 | __q08, __q09, __q10, __q11, __q12, __q13, __q14, __q15, | |
1306 | __q16, __q17, __q18, __q19, __q20, __q21, __q22, __q23, | |
1307 | __q24, __q25, __q26, __q27, __q28, __q29, __q30, __q31 | |
1308 | }; | |
1309 | } | |
1310 | ||
1311 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1312 | _mm256_set_epi64x (long long __A, long long __B, long long __C, | |
1313 | long long __D) | |
1314 | { | |
1315 | return __extension__ (__m256i)(__v4di){ __D, __C, __B, __A }; | |
1316 | } | |
1317 | ||
1318 | /* Create a vector with all elements equal to A. */ | |
1319 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1320 | _mm256_set1_pd (double __A) | |
1321 | { | |
1322 | return __extension__ (__m256d){ __A, __A, __A, __A }; | |
1323 | } | |
1324 | ||
1325 | /* Create a vector with all elements equal to A. */ | |
1326 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1327 | _mm256_set1_ps (float __A) | |
1328 | { | |
1329 | return __extension__ (__m256){ __A, __A, __A, __A, | |
1330 | __A, __A, __A, __A }; | |
1331 | } | |
1332 | ||
1333 | /* Create a vector with all elements equal to A. */ | |
1334 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1335 | _mm256_set1_epi32 (int __A) | |
1336 | { | |
1337 | return __extension__ (__m256i)(__v8si){ __A, __A, __A, __A, | |
1338 | __A, __A, __A, __A }; | |
1339 | } | |
1340 | ||
1341 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1342 | _mm256_set1_epi16 (short __A) | |
1343 | { | |
1344 | return _mm256_set_epi16 (__A, __A, __A, __A, __A, __A, __A, __A, | |
1345 | __A, __A, __A, __A, __A, __A, __A, __A); | |
1346 | } | |
1347 | ||
1348 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1349 | _mm256_set1_epi8 (char __A) | |
1350 | { | |
1351 | return _mm256_set_epi8 (__A, __A, __A, __A, __A, __A, __A, __A, | |
1352 | __A, __A, __A, __A, __A, __A, __A, __A, | |
1353 | __A, __A, __A, __A, __A, __A, __A, __A, | |
1354 | __A, __A, __A, __A, __A, __A, __A, __A); | |
1355 | } | |
1356 | ||
1357 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1358 | _mm256_set1_epi64x (long long __A) | |
1359 | { | |
1360 | return __extension__ (__m256i)(__v4di){ __A, __A, __A, __A }; | |
1361 | } | |
1362 | ||
1363 | /* Create vectors of elements in the reversed order from the | |
1364 | _mm256_set_XXX functions. */ | |
1365 | ||
1366 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1367 | _mm256_setr_pd (double __A, double __B, double __C, double __D) | |
1368 | { | |
1369 | return _mm256_set_pd (__D, __C, __B, __A); | |
1370 | } | |
1371 | ||
1372 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1373 | _mm256_setr_ps (float __A, float __B, float __C, float __D, | |
1374 | float __E, float __F, float __G, float __H) | |
1375 | { | |
1376 | return _mm256_set_ps (__H, __G, __F, __E, __D, __C, __B, __A); | |
1377 | } | |
1378 | ||
1379 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1380 | _mm256_setr_epi32 (int __A, int __B, int __C, int __D, | |
1381 | int __E, int __F, int __G, int __H) | |
1382 | { | |
1383 | return _mm256_set_epi32 (__H, __G, __F, __E, __D, __C, __B, __A); | |
1384 | } | |
1385 | ||
1386 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1387 | _mm256_setr_epi16 (short __q15, short __q14, short __q13, short __q12, | |
1388 | short __q11, short __q10, short __q09, short __q08, | |
1389 | short __q07, short __q06, short __q05, short __q04, | |
1390 | short __q03, short __q02, short __q01, short __q00) | |
1391 | { | |
1392 | return _mm256_set_epi16 (__q00, __q01, __q02, __q03, | |
1393 | __q04, __q05, __q06, __q07, | |
1394 | __q08, __q09, __q10, __q11, | |
1395 | __q12, __q13, __q14, __q15); | |
1396 | } | |
1397 | ||
1398 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1399 | _mm256_setr_epi8 (char __q31, char __q30, char __q29, char __q28, | |
1400 | char __q27, char __q26, char __q25, char __q24, | |
1401 | char __q23, char __q22, char __q21, char __q20, | |
1402 | char __q19, char __q18, char __q17, char __q16, | |
1403 | char __q15, char __q14, char __q13, char __q12, | |
1404 | char __q11, char __q10, char __q09, char __q08, | |
1405 | char __q07, char __q06, char __q05, char __q04, | |
1406 | char __q03, char __q02, char __q01, char __q00) | |
1407 | { | |
1408 | return _mm256_set_epi8 (__q00, __q01, __q02, __q03, | |
1409 | __q04, __q05, __q06, __q07, | |
1410 | __q08, __q09, __q10, __q11, | |
1411 | __q12, __q13, __q14, __q15, | |
1412 | __q16, __q17, __q18, __q19, | |
1413 | __q20, __q21, __q22, __q23, | |
1414 | __q24, __q25, __q26, __q27, | |
1415 | __q28, __q29, __q30, __q31); | |
1416 | } | |
1417 | ||
1418 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1419 | _mm256_setr_epi64x (long long __A, long long __B, long long __C, | |
1420 | long long __D) | |
1421 | { | |
1422 | return _mm256_set_epi64x (__D, __C, __B, __A); | |
1423 | } | |
1424 | ||
1425 | /* Casts between various SP, DP, INT vector types. Note that these do no | |
1426 | conversion of values, they just change the type. */ | |
1427 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1428 | _mm256_castpd_ps (__m256d __A) | |
1429 | { | |
1430 | return (__m256) __A; | |
1431 | } | |
1432 | ||
1433 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1434 | _mm256_castpd_si256 (__m256d __A) | |
1435 | { | |
1436 | return (__m256i) __A; | |
1437 | } | |
1438 | ||
1439 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1440 | _mm256_castps_pd (__m256 __A) | |
1441 | { | |
1442 | return (__m256d) __A; | |
1443 | } | |
1444 | ||
1445 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1446 | _mm256_castps_si256(__m256 __A) | |
1447 | { | |
1448 | return (__m256i) __A; | |
1449 | } | |
1450 | ||
1451 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1452 | _mm256_castsi256_ps (__m256i __A) | |
1453 | { | |
1454 | return (__m256) __A; | |
1455 | } | |
1456 | ||
1457 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1458 | _mm256_castsi256_pd (__m256i __A) | |
1459 | { | |
1460 | return (__m256d) __A; | |
1461 | } | |
1462 | ||
1463 | extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1464 | _mm256_castpd256_pd128 (__m256d __A) | |
1465 | { | |
1466 | return (__m128d) __builtin_ia32_pd_pd256 ((__v4df)__A); | |
1467 | } | |
1468 | ||
1469 | extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1470 | _mm256_castps256_ps128 (__m256 __A) | |
1471 | { | |
1472 | return (__m128) __builtin_ia32_ps_ps256 ((__v8sf)__A); | |
1473 | } | |
1474 | ||
1475 | extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1476 | _mm256_castsi256_si128 (__m256i __A) | |
1477 | { | |
1478 | return (__m128i) __builtin_ia32_si_si256 ((__v8si)__A); | |
1479 | } | |
1480 | ||
1481 | /* When cast is done from a 128 to 256-bit type, the low 128 bits of | |
1482 | the 256-bit result contain source parameter value and the upper 128 | |
1483 | bits of the result are undefined. Those intrinsics shouldn't | |
1484 | generate any extra moves. */ | |
1485 | ||
1486 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1487 | _mm256_castpd128_pd256 (__m128d __A) | |
1488 | { | |
1489 | return (__m256d) __builtin_ia32_pd256_pd ((__v2df)__A); | |
1490 | } | |
1491 | ||
1492 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1493 | _mm256_castps128_ps256 (__m128 __A) | |
1494 | { | |
1495 | return (__m256) __builtin_ia32_ps256_ps ((__v4sf)__A); | |
1496 | } | |
1497 | ||
1498 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1499 | _mm256_castsi128_si256 (__m128i __A) | |
1500 | { | |
1501 | return (__m256i) __builtin_ia32_si256_si ((__v4si)__A); | |
1502 | } | |
97db2bf7 | 1503 | |
e6b2dc24 JJ |
1504 | /* Similarly, but with zero extension instead of undefined values. */ |
1505 | ||
1506 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1507 | _mm256_zextpd128_pd256 (__m128d __A) | |
1508 | { | |
1509 | return _mm256_insertf128_pd (_mm256_setzero_pd (), __A, 0); | |
1510 | } | |
1511 | ||
1512 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1513 | _mm256_zextps128_ps256 (__m128 __A) | |
1514 | { | |
1515 | return _mm256_insertf128_ps (_mm256_setzero_ps (), __A, 0); | |
1516 | } | |
1517 | ||
1518 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1519 | _mm256_zextsi128_si256 (__m128i __A) | |
1520 | { | |
1521 | return _mm256_insertf128_si256 (_mm256_setzero_si256 (), __A, 0); | |
1522 | } | |
1523 | ||
f4ee3a9e UB |
1524 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
1525 | _mm256_set_m128 ( __m128 __H, __m128 __L) | |
1526 | { | |
1527 | return _mm256_insertf128_ps (_mm256_castps128_ps256 (__L), __H, 1); | |
1528 | } | |
1529 | ||
1530 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1531 | _mm256_set_m128d (__m128d __H, __m128d __L) | |
1532 | { | |
1533 | return _mm256_insertf128_pd (_mm256_castpd128_pd256 (__L), __H, 1); | |
1534 | } | |
1535 | ||
1536 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1537 | _mm256_set_m128i (__m128i __H, __m128i __L) | |
1538 | { | |
1539 | return _mm256_insertf128_si256 (_mm256_castsi128_si256 (__L), __H, 1); | |
1540 | } | |
1541 | ||
1542 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1543 | _mm256_setr_m128 (__m128 __L, __m128 __H) | |
1544 | { | |
1545 | return _mm256_set_m128 (__H, __L); | |
1546 | } | |
1547 | ||
1548 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1549 | _mm256_setr_m128d (__m128d __L, __m128d __H) | |
1550 | { | |
1551 | return _mm256_set_m128d (__H, __L); | |
1552 | } | |
1553 | ||
1554 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1555 | _mm256_setr_m128i (__m128i __L, __m128i __H) | |
1556 | { | |
1557 | return _mm256_set_m128i (__H, __L); | |
1558 | } | |
1559 | ||
96d5c6dc JJ |
1560 | extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
1561 | _mm256_loadu2_m128 (float const *__PH, float const *__PL) | |
1562 | { | |
1563 | return _mm256_insertf128_ps (_mm256_castps128_ps256 (_mm_loadu_ps (__PL)), | |
1564 | _mm_loadu_ps (__PH), 1); | |
1565 | } | |
1566 | ||
1567 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1568 | _mm256_storeu2_m128 (float *__PH, float *__PL, __m256 __A) | |
1569 | { | |
1570 | _mm_storeu_ps (__PL, _mm256_castps256_ps128 (__A)); | |
1571 | _mm_storeu_ps (__PH, _mm256_extractf128_ps (__A, 1)); | |
1572 | } | |
1573 | ||
1574 | extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1575 | _mm256_loadu2_m128d (double const *__PH, double const *__PL) | |
1576 | { | |
1577 | return _mm256_insertf128_pd (_mm256_castpd128_pd256 (_mm_loadu_pd (__PL)), | |
1578 | _mm_loadu_pd (__PH), 1); | |
1579 | } | |
1580 | ||
1581 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1582 | _mm256_storeu2_m128d (double *__PH, double *__PL, __m256d __A) | |
1583 | { | |
1584 | _mm_storeu_pd (__PL, _mm256_castpd256_pd128 (__A)); | |
1585 | _mm_storeu_pd (__PH, _mm256_extractf128_pd (__A, 1)); | |
1586 | } | |
1587 | ||
1588 | extern __inline __m256i __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1589 | _mm256_loadu2_m128i (__m128i_u const *__PH, __m128i_u const *__PL) | |
1590 | { | |
1591 | return _mm256_insertf128_si256 (_mm256_castsi128_si256 (_mm_loadu_si128 (__PL)), | |
1592 | _mm_loadu_si128 (__PH), 1); | |
1593 | } | |
1594 | ||
1595 | extern __inline void __attribute__((__gnu_inline__, __always_inline__, __artificial__)) | |
1596 | _mm256_storeu2_m128i (__m128i_u *__PH, __m128i_u *__PL, __m256i __A) | |
1597 | { | |
1598 | _mm_storeu_si128 (__PL, _mm256_castsi256_si128 (__A)); | |
1599 | _mm_storeu_si128 (__PH, _mm256_extractf128_si256 (__A, 1)); | |
1600 | } | |
1601 | ||
97db2bf7 ST |
1602 | #ifdef __DISABLE_AVX__ |
1603 | #undef __DISABLE_AVX__ | |
1604 | #pragma GCC pop_options | |
1605 | #endif /* __DISABLE_AVX__ */ | |
1606 | ||
1607 | #endif /* _AVXINTRIN_H_INCLUDED */ |