xref: /freebsd/contrib/llvm-project/clang/lib/Headers/avxintrin.h (revision 6580f5c38dd5b01aeeaed16b370f1a12423437f0)
1 /*===---- avxintrin.h - AVX intrinsics -------------------------------------===
2  *
3  * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4  * See https://llvm.org/LICENSE.txt for license information.
5  * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6  *
7  *===-----------------------------------------------------------------------===
8  */
9 
10 #ifndef __IMMINTRIN_H
11 #error "Never use <avxintrin.h> directly; include <immintrin.h> instead."
12 #endif
13 
14 #ifndef __AVXINTRIN_H
15 #define __AVXINTRIN_H
16 
17 typedef double __v4df __attribute__ ((__vector_size__ (32)));
18 typedef float __v8sf __attribute__ ((__vector_size__ (32)));
19 typedef long long __v4di __attribute__ ((__vector_size__ (32)));
20 typedef int __v8si __attribute__ ((__vector_size__ (32)));
21 typedef short __v16hi __attribute__ ((__vector_size__ (32)));
22 typedef char __v32qi __attribute__ ((__vector_size__ (32)));
23 
24 /* Unsigned types */
25 typedef unsigned long long __v4du __attribute__ ((__vector_size__ (32)));
26 typedef unsigned int __v8su __attribute__ ((__vector_size__ (32)));
27 typedef unsigned short __v16hu __attribute__ ((__vector_size__ (32)));
28 typedef unsigned char __v32qu __attribute__ ((__vector_size__ (32)));
29 
30 /* We need an explicitly signed variant for char. Note that this shouldn't
31  * appear in the interface though. */
32 typedef signed char __v32qs __attribute__((__vector_size__(32)));
33 
34 typedef float __m256 __attribute__ ((__vector_size__ (32), __aligned__(32)));
35 typedef double __m256d __attribute__((__vector_size__(32), __aligned__(32)));
36 typedef long long __m256i __attribute__((__vector_size__(32), __aligned__(32)));
37 
38 typedef float __m256_u __attribute__ ((__vector_size__ (32), __aligned__(1)));
39 typedef double __m256d_u __attribute__((__vector_size__(32), __aligned__(1)));
40 typedef long long __m256i_u __attribute__((__vector_size__(32), __aligned__(1)));
41 
42 #ifdef __SSE2__
43 /* Both _Float16 and __bf16 require SSE2 being enabled. */
44 typedef _Float16 __v16hf __attribute__((__vector_size__(32), __aligned__(32)));
45 typedef _Float16 __m256h __attribute__((__vector_size__(32), __aligned__(32)));
46 typedef _Float16 __m256h_u __attribute__((__vector_size__(32), __aligned__(1)));
47 
48 typedef __bf16 __v16bf __attribute__((__vector_size__(32), __aligned__(32)));
49 typedef __bf16 __m256bh __attribute__((__vector_size__(32), __aligned__(32)));
50 #endif
51 
52 /* Define the default attributes for the functions in this file. */
53 #define __DEFAULT_FN_ATTRS                                                     \
54   __attribute__((__always_inline__, __nodebug__, __target__("avx,no-evex512"), \
55                  __min_vector_width__(256)))
56 #define __DEFAULT_FN_ATTRS128                                                  \
57   __attribute__((__always_inline__, __nodebug__, __target__("avx,no-evex512"), \
58                  __min_vector_width__(128)))
59 
60 /* Arithmetic */
61 /// Adds two 256-bit vectors of [4 x double].
62 ///
63 /// \headerfile <x86intrin.h>
64 ///
65 /// This intrinsic corresponds to the <c> VADDPD </c> instruction.
66 ///
67 /// \param __a
68 ///    A 256-bit vector of [4 x double] containing one of the source operands.
69 /// \param __b
70 ///    A 256-bit vector of [4 x double] containing one of the source operands.
71 /// \returns A 256-bit vector of [4 x double] containing the sums of both
72 ///    operands.
73 static __inline __m256d __DEFAULT_FN_ATTRS
74 _mm256_add_pd(__m256d __a, __m256d __b)
75 {
76   return (__m256d)((__v4df)__a+(__v4df)__b);
77 }
78 
79 /// Adds two 256-bit vectors of [8 x float].
80 ///
81 /// \headerfile <x86intrin.h>
82 ///
83 /// This intrinsic corresponds to the <c> VADDPS </c> instruction.
84 ///
85 /// \param __a
86 ///    A 256-bit vector of [8 x float] containing one of the source operands.
87 /// \param __b
88 ///    A 256-bit vector of [8 x float] containing one of the source operands.
89 /// \returns A 256-bit vector of [8 x float] containing the sums of both
90 ///    operands.
91 static __inline __m256 __DEFAULT_FN_ATTRS
92 _mm256_add_ps(__m256 __a, __m256 __b)
93 {
94   return (__m256)((__v8sf)__a+(__v8sf)__b);
95 }
96 
97 /// Subtracts two 256-bit vectors of [4 x double].
98 ///
99 /// \headerfile <x86intrin.h>
100 ///
101 /// This intrinsic corresponds to the <c> VSUBPD </c> instruction.
102 ///
103 /// \param __a
104 ///    A 256-bit vector of [4 x double] containing the minuend.
105 /// \param __b
106 ///    A 256-bit vector of [4 x double] containing the subtrahend.
107 /// \returns A 256-bit vector of [4 x double] containing the differences between
108 ///    both operands.
109 static __inline __m256d __DEFAULT_FN_ATTRS
110 _mm256_sub_pd(__m256d __a, __m256d __b)
111 {
112   return (__m256d)((__v4df)__a-(__v4df)__b);
113 }
114 
115 /// Subtracts two 256-bit vectors of [8 x float].
116 ///
117 /// \headerfile <x86intrin.h>
118 ///
119 /// This intrinsic corresponds to the <c> VSUBPS </c> instruction.
120 ///
121 /// \param __a
122 ///    A 256-bit vector of [8 x float] containing the minuend.
123 /// \param __b
124 ///    A 256-bit vector of [8 x float] containing the subtrahend.
125 /// \returns A 256-bit vector of [8 x float] containing the differences between
126 ///    both operands.
127 static __inline __m256 __DEFAULT_FN_ATTRS
128 _mm256_sub_ps(__m256 __a, __m256 __b)
129 {
130   return (__m256)((__v8sf)__a-(__v8sf)__b);
131 }
132 
133 /// Adds the even-indexed values and subtracts the odd-indexed values of
134 ///    two 256-bit vectors of [4 x double].
135 ///
136 /// \headerfile <x86intrin.h>
137 ///
138 /// This intrinsic corresponds to the <c> VADDSUBPD </c> instruction.
139 ///
140 /// \param __a
141 ///    A 256-bit vector of [4 x double] containing the left source operand.
142 /// \param __b
143 ///    A 256-bit vector of [4 x double] containing the right source operand.
144 /// \returns A 256-bit vector of [4 x double] containing the alternating sums
145 ///    and differences between both operands.
146 static __inline __m256d __DEFAULT_FN_ATTRS
147 _mm256_addsub_pd(__m256d __a, __m256d __b)
148 {
149   return (__m256d)__builtin_ia32_addsubpd256((__v4df)__a, (__v4df)__b);
150 }
151 
152 /// Adds the even-indexed values and subtracts the odd-indexed values of
153 ///    two 256-bit vectors of [8 x float].
154 ///
155 /// \headerfile <x86intrin.h>
156 ///
157 /// This intrinsic corresponds to the <c> VADDSUBPS </c> instruction.
158 ///
159 /// \param __a
160 ///    A 256-bit vector of [8 x float] containing the left source operand.
161 /// \param __b
162 ///    A 256-bit vector of [8 x float] containing the right source operand.
163 /// \returns A 256-bit vector of [8 x float] containing the alternating sums and
164 ///    differences between both operands.
165 static __inline __m256 __DEFAULT_FN_ATTRS
166 _mm256_addsub_ps(__m256 __a, __m256 __b)
167 {
168   return (__m256)__builtin_ia32_addsubps256((__v8sf)__a, (__v8sf)__b);
169 }
170 
171 /// Divides two 256-bit vectors of [4 x double].
172 ///
173 /// \headerfile <x86intrin.h>
174 ///
175 /// This intrinsic corresponds to the <c> VDIVPD </c> instruction.
176 ///
177 /// \param __a
178 ///    A 256-bit vector of [4 x double] containing the dividend.
179 /// \param __b
180 ///    A 256-bit vector of [4 x double] containing the divisor.
181 /// \returns A 256-bit vector of [4 x double] containing the quotients of both
182 ///    operands.
183 static __inline __m256d __DEFAULT_FN_ATTRS
184 _mm256_div_pd(__m256d __a, __m256d __b)
185 {
186   return (__m256d)((__v4df)__a/(__v4df)__b);
187 }
188 
189 /// Divides two 256-bit vectors of [8 x float].
190 ///
191 /// \headerfile <x86intrin.h>
192 ///
193 /// This intrinsic corresponds to the <c> VDIVPS </c> instruction.
194 ///
195 /// \param __a
196 ///    A 256-bit vector of [8 x float] containing the dividend.
197 /// \param __b
198 ///    A 256-bit vector of [8 x float] containing the divisor.
199 /// \returns A 256-bit vector of [8 x float] containing the quotients of both
200 ///    operands.
201 static __inline __m256 __DEFAULT_FN_ATTRS
202 _mm256_div_ps(__m256 __a, __m256 __b)
203 {
204   return (__m256)((__v8sf)__a/(__v8sf)__b);
205 }
206 
207 /// Compares two 256-bit vectors of [4 x double] and returns the greater
208 ///    of each pair of values.
209 ///
210 /// \headerfile <x86intrin.h>
211 ///
212 /// This intrinsic corresponds to the <c> VMAXPD </c> instruction.
213 ///
214 /// \param __a
215 ///    A 256-bit vector of [4 x double] containing one of the operands.
216 /// \param __b
217 ///    A 256-bit vector of [4 x double] containing one of the operands.
218 /// \returns A 256-bit vector of [4 x double] containing the maximum values
219 ///    between both operands.
220 static __inline __m256d __DEFAULT_FN_ATTRS
221 _mm256_max_pd(__m256d __a, __m256d __b)
222 {
223   return (__m256d)__builtin_ia32_maxpd256((__v4df)__a, (__v4df)__b);
224 }
225 
226 /// Compares two 256-bit vectors of [8 x float] and returns the greater
227 ///    of each pair of values.
228 ///
229 /// \headerfile <x86intrin.h>
230 ///
231 /// This intrinsic corresponds to the <c> VMAXPS </c> instruction.
232 ///
233 /// \param __a
234 ///    A 256-bit vector of [8 x float] containing one of the operands.
235 /// \param __b
236 ///    A 256-bit vector of [8 x float] containing one of the operands.
237 /// \returns A 256-bit vector of [8 x float] containing the maximum values
238 ///    between both operands.
239 static __inline __m256 __DEFAULT_FN_ATTRS
240 _mm256_max_ps(__m256 __a, __m256 __b)
241 {
242   return (__m256)__builtin_ia32_maxps256((__v8sf)__a, (__v8sf)__b);
243 }
244 
245 /// Compares two 256-bit vectors of [4 x double] and returns the lesser
246 ///    of each pair of values.
247 ///
248 /// \headerfile <x86intrin.h>
249 ///
250 /// This intrinsic corresponds to the <c> VMINPD </c> instruction.
251 ///
252 /// \param __a
253 ///    A 256-bit vector of [4 x double] containing one of the operands.
254 /// \param __b
255 ///    A 256-bit vector of [4 x double] containing one of the operands.
256 /// \returns A 256-bit vector of [4 x double] containing the minimum values
257 ///    between both operands.
258 static __inline __m256d __DEFAULT_FN_ATTRS
259 _mm256_min_pd(__m256d __a, __m256d __b)
260 {
261   return (__m256d)__builtin_ia32_minpd256((__v4df)__a, (__v4df)__b);
262 }
263 
264 /// Compares two 256-bit vectors of [8 x float] and returns the lesser
265 ///    of each pair of values.
266 ///
267 /// \headerfile <x86intrin.h>
268 ///
269 /// This intrinsic corresponds to the <c> VMINPS </c> instruction.
270 ///
271 /// \param __a
272 ///    A 256-bit vector of [8 x float] containing one of the operands.
273 /// \param __b
274 ///    A 256-bit vector of [8 x float] containing one of the operands.
275 /// \returns A 256-bit vector of [8 x float] containing the minimum values
276 ///    between both operands.
277 static __inline __m256 __DEFAULT_FN_ATTRS
278 _mm256_min_ps(__m256 __a, __m256 __b)
279 {
280   return (__m256)__builtin_ia32_minps256((__v8sf)__a, (__v8sf)__b);
281 }
282 
283 /// Multiplies two 256-bit vectors of [4 x double].
284 ///
285 /// \headerfile <x86intrin.h>
286 ///
287 /// This intrinsic corresponds to the <c> VMULPD </c> instruction.
288 ///
289 /// \param __a
290 ///    A 256-bit vector of [4 x double] containing one of the operands.
291 /// \param __b
292 ///    A 256-bit vector of [4 x double] containing one of the operands.
293 /// \returns A 256-bit vector of [4 x double] containing the products of both
294 ///    operands.
295 static __inline __m256d __DEFAULT_FN_ATTRS
296 _mm256_mul_pd(__m256d __a, __m256d __b)
297 {
298   return (__m256d)((__v4df)__a * (__v4df)__b);
299 }
300 
301 /// Multiplies two 256-bit vectors of [8 x float].
302 ///
303 /// \headerfile <x86intrin.h>
304 ///
305 /// This intrinsic corresponds to the <c> VMULPS </c> instruction.
306 ///
307 /// \param __a
308 ///    A 256-bit vector of [8 x float] containing one of the operands.
309 /// \param __b
310 ///    A 256-bit vector of [8 x float] containing one of the operands.
311 /// \returns A 256-bit vector of [8 x float] containing the products of both
312 ///    operands.
313 static __inline __m256 __DEFAULT_FN_ATTRS
314 _mm256_mul_ps(__m256 __a, __m256 __b)
315 {
316   return (__m256)((__v8sf)__a * (__v8sf)__b);
317 }
318 
319 /// Calculates the square roots of the values in a 256-bit vector of
320 ///    [4 x double].
321 ///
322 /// \headerfile <x86intrin.h>
323 ///
324 /// This intrinsic corresponds to the <c> VSQRTPD </c> instruction.
325 ///
326 /// \param __a
327 ///    A 256-bit vector of [4 x double].
328 /// \returns A 256-bit vector of [4 x double] containing the square roots of the
329 ///    values in the operand.
330 static __inline __m256d __DEFAULT_FN_ATTRS
331 _mm256_sqrt_pd(__m256d __a)
332 {
333   return (__m256d)__builtin_ia32_sqrtpd256((__v4df)__a);
334 }
335 
336 /// Calculates the square roots of the values in a 256-bit vector of
337 ///    [8 x float].
338 ///
339 /// \headerfile <x86intrin.h>
340 ///
341 /// This intrinsic corresponds to the <c> VSQRTPS </c> instruction.
342 ///
343 /// \param __a
344 ///    A 256-bit vector of [8 x float].
345 /// \returns A 256-bit vector of [8 x float] containing the square roots of the
346 ///    values in the operand.
347 static __inline __m256 __DEFAULT_FN_ATTRS
348 _mm256_sqrt_ps(__m256 __a)
349 {
350   return (__m256)__builtin_ia32_sqrtps256((__v8sf)__a);
351 }
352 
353 /// Calculates the reciprocal square roots of the values in a 256-bit
354 ///    vector of [8 x float].
355 ///
356 /// \headerfile <x86intrin.h>
357 ///
358 /// This intrinsic corresponds to the <c> VRSQRTPS </c> instruction.
359 ///
360 /// \param __a
361 ///    A 256-bit vector of [8 x float].
362 /// \returns A 256-bit vector of [8 x float] containing the reciprocal square
363 ///    roots of the values in the operand.
364 static __inline __m256 __DEFAULT_FN_ATTRS
365 _mm256_rsqrt_ps(__m256 __a)
366 {
367   return (__m256)__builtin_ia32_rsqrtps256((__v8sf)__a);
368 }
369 
370 /// Calculates the reciprocals of the values in a 256-bit vector of
371 ///    [8 x float].
372 ///
373 /// \headerfile <x86intrin.h>
374 ///
375 /// This intrinsic corresponds to the <c> VRCPPS </c> instruction.
376 ///
377 /// \param __a
378 ///    A 256-bit vector of [8 x float].
379 /// \returns A 256-bit vector of [8 x float] containing the reciprocals of the
380 ///    values in the operand.
381 static __inline __m256 __DEFAULT_FN_ATTRS
382 _mm256_rcp_ps(__m256 __a)
383 {
384   return (__m256)__builtin_ia32_rcpps256((__v8sf)__a);
385 }
386 
387 /// Rounds the values in a 256-bit vector of [4 x double] as specified
388 ///    by the byte operand. The source values are rounded to integer values and
389 ///    returned as 64-bit double-precision floating-point values.
390 ///
391 /// \headerfile <x86intrin.h>
392 ///
393 /// \code
394 /// __m256d _mm256_round_pd(__m256d V, const int M);
395 /// \endcode
396 ///
397 /// This intrinsic corresponds to the <c> VROUNDPD </c> instruction.
398 ///
399 /// \param V
400 ///    A 256-bit vector of [4 x double].
401 /// \param M
402 ///    An integer value that specifies the rounding operation. \n
403 ///    Bits [7:4] are reserved. \n
404 ///    Bit [3] is a precision exception value: \n
405 ///      0: A normal PE exception is used. \n
406 ///      1: The PE field is not updated. \n
407 ///    Bit [2] is the rounding control source: \n
408 ///      0: Use bits [1:0] of \a M. \n
409 ///      1: Use the current MXCSR setting. \n
410 ///    Bits [1:0] contain the rounding control definition: \n
411 ///      00: Nearest. \n
412 ///      01: Downward (toward negative infinity). \n
413 ///      10: Upward (toward positive infinity). \n
414 ///      11: Truncated.
415 /// \returns A 256-bit vector of [4 x double] containing the rounded values.
416 #define _mm256_round_pd(V, M) \
417   ((__m256d)__builtin_ia32_roundpd256((__v4df)(__m256d)(V), (M)))
418 
419 /// Rounds the values stored in a 256-bit vector of [8 x float] as
420 ///    specified by the byte operand. The source values are rounded to integer
421 ///    values and returned as floating-point values.
422 ///
423 /// \headerfile <x86intrin.h>
424 ///
425 /// \code
426 /// __m256 _mm256_round_ps(__m256 V, const int M);
427 /// \endcode
428 ///
429 /// This intrinsic corresponds to the <c> VROUNDPS </c> instruction.
430 ///
431 /// \param V
432 ///    A 256-bit vector of [8 x float].
433 /// \param M
434 ///    An integer value that specifies the rounding operation. \n
435 ///    Bits [7:4] are reserved. \n
436 ///    Bit [3] is a precision exception value: \n
437 ///      0: A normal PE exception is used. \n
438 ///      1: The PE field is not updated. \n
439 ///    Bit [2] is the rounding control source: \n
440 ///      0: Use bits [1:0] of \a M. \n
441 ///      1: Use the current MXCSR setting. \n
442 ///    Bits [1:0] contain the rounding control definition: \n
443 ///      00: Nearest. \n
444 ///      01: Downward (toward negative infinity). \n
445 ///      10: Upward (toward positive infinity). \n
446 ///      11: Truncated.
447 /// \returns A 256-bit vector of [8 x float] containing the rounded values.
448 #define _mm256_round_ps(V, M) \
449   ((__m256)__builtin_ia32_roundps256((__v8sf)(__m256)(V), (M)))
450 
451 /// Rounds up the values stored in a 256-bit vector of [4 x double]. The
452 ///    source values are rounded up to integer values and returned as 64-bit
453 ///    double-precision floating-point values.
454 ///
455 /// \headerfile <x86intrin.h>
456 ///
457 /// \code
458 /// __m256d _mm256_ceil_pd(__m256d V);
459 /// \endcode
460 ///
461 /// This intrinsic corresponds to the <c> VROUNDPD </c> instruction.
462 ///
463 /// \param V
464 ///    A 256-bit vector of [4 x double].
465 /// \returns A 256-bit vector of [4 x double] containing the rounded up values.
466 #define _mm256_ceil_pd(V)  _mm256_round_pd((V), _MM_FROUND_CEIL)
467 
468 /// Rounds down the values stored in a 256-bit vector of [4 x double].
469 ///    The source values are rounded down to integer values and returned as
470 ///    64-bit double-precision floating-point values.
471 ///
472 /// \headerfile <x86intrin.h>
473 ///
474 /// \code
475 /// __m256d _mm256_floor_pd(__m256d V);
476 /// \endcode
477 ///
478 /// This intrinsic corresponds to the <c> VROUNDPD </c> instruction.
479 ///
480 /// \param V
481 ///    A 256-bit vector of [4 x double].
482 /// \returns A 256-bit vector of [4 x double] containing the rounded down
483 ///    values.
484 #define _mm256_floor_pd(V) _mm256_round_pd((V), _MM_FROUND_FLOOR)
485 
486 /// Rounds up the values stored in a 256-bit vector of [8 x float]. The
487 ///    source values are rounded up to integer values and returned as
488 ///    floating-point values.
489 ///
490 /// \headerfile <x86intrin.h>
491 ///
492 /// \code
493 /// __m256 _mm256_ceil_ps(__m256 V);
494 /// \endcode
495 ///
496 /// This intrinsic corresponds to the <c> VROUNDPS </c> instruction.
497 ///
498 /// \param V
499 ///    A 256-bit vector of [8 x float].
500 /// \returns A 256-bit vector of [8 x float] containing the rounded up values.
501 #define _mm256_ceil_ps(V)  _mm256_round_ps((V), _MM_FROUND_CEIL)
502 
503 /// Rounds down the values stored in a 256-bit vector of [8 x float]. The
504 ///    source values are rounded down to integer values and returned as
505 ///    floating-point values.
506 ///
507 /// \headerfile <x86intrin.h>
508 ///
509 /// \code
510 /// __m256 _mm256_floor_ps(__m256 V);
511 /// \endcode
512 ///
513 /// This intrinsic corresponds to the <c> VROUNDPS </c> instruction.
514 ///
515 /// \param V
516 ///    A 256-bit vector of [8 x float].
517 /// \returns A 256-bit vector of [8 x float] containing the rounded down values.
518 #define _mm256_floor_ps(V) _mm256_round_ps((V), _MM_FROUND_FLOOR)
519 
520 /* Logical */
521 /// Performs a bitwise AND of two 256-bit vectors of [4 x double].
522 ///
523 /// \headerfile <x86intrin.h>
524 ///
525 /// This intrinsic corresponds to the <c> VANDPD </c> instruction.
526 ///
527 /// \param __a
528 ///    A 256-bit vector of [4 x double] containing one of the source operands.
529 /// \param __b
530 ///    A 256-bit vector of [4 x double] containing one of the source operands.
531 /// \returns A 256-bit vector of [4 x double] containing the bitwise AND of the
532 ///    values between both operands.
533 static __inline __m256d __DEFAULT_FN_ATTRS
534 _mm256_and_pd(__m256d __a, __m256d __b)
535 {
536   return (__m256d)((__v4du)__a & (__v4du)__b);
537 }
538 
539 /// Performs a bitwise AND of two 256-bit vectors of [8 x float].
540 ///
541 /// \headerfile <x86intrin.h>
542 ///
543 /// This intrinsic corresponds to the <c> VANDPS </c> instruction.
544 ///
545 /// \param __a
546 ///    A 256-bit vector of [8 x float] containing one of the source operands.
547 /// \param __b
548 ///    A 256-bit vector of [8 x float] containing one of the source operands.
549 /// \returns A 256-bit vector of [8 x float] containing the bitwise AND of the
550 ///    values between both operands.
551 static __inline __m256 __DEFAULT_FN_ATTRS
552 _mm256_and_ps(__m256 __a, __m256 __b)
553 {
554   return (__m256)((__v8su)__a & (__v8su)__b);
555 }
556 
557 /// Performs a bitwise AND of two 256-bit vectors of [4 x double], using
558 ///    the one's complement of the values contained in the first source operand.
559 ///
560 /// \headerfile <x86intrin.h>
561 ///
562 /// This intrinsic corresponds to the <c> VANDNPD </c> instruction.
563 ///
564 /// \param __a
565 ///    A 256-bit vector of [4 x double] containing the left source operand. The
566 ///    one's complement of this value is used in the bitwise AND.
567 /// \param __b
568 ///    A 256-bit vector of [4 x double] containing the right source operand.
569 /// \returns A 256-bit vector of [4 x double] containing the bitwise AND of the
570 ///    values of the second operand and the one's complement of the first
571 ///    operand.
572 static __inline __m256d __DEFAULT_FN_ATTRS
573 _mm256_andnot_pd(__m256d __a, __m256d __b)
574 {
575   return (__m256d)(~(__v4du)__a & (__v4du)__b);
576 }
577 
578 /// Performs a bitwise AND of two 256-bit vectors of [8 x float], using
579 ///    the one's complement of the values contained in the first source operand.
580 ///
581 /// \headerfile <x86intrin.h>
582 ///
583 /// This intrinsic corresponds to the <c> VANDNPS </c> instruction.
584 ///
585 /// \param __a
586 ///    A 256-bit vector of [8 x float] containing the left source operand. The
587 ///    one's complement of this value is used in the bitwise AND.
588 /// \param __b
589 ///    A 256-bit vector of [8 x float] containing the right source operand.
590 /// \returns A 256-bit vector of [8 x float] containing the bitwise AND of the
591 ///    values of the second operand and the one's complement of the first
592 ///    operand.
593 static __inline __m256 __DEFAULT_FN_ATTRS
594 _mm256_andnot_ps(__m256 __a, __m256 __b)
595 {
596   return (__m256)(~(__v8su)__a & (__v8su)__b);
597 }
598 
599 /// Performs a bitwise OR of two 256-bit vectors of [4 x double].
600 ///
601 /// \headerfile <x86intrin.h>
602 ///
603 /// This intrinsic corresponds to the <c> VORPD </c> instruction.
604 ///
605 /// \param __a
606 ///    A 256-bit vector of [4 x double] containing one of the source operands.
607 /// \param __b
608 ///    A 256-bit vector of [4 x double] containing one of the source operands.
609 /// \returns A 256-bit vector of [4 x double] containing the bitwise OR of the
610 ///    values between both operands.
611 static __inline __m256d __DEFAULT_FN_ATTRS
612 _mm256_or_pd(__m256d __a, __m256d __b)
613 {
614   return (__m256d)((__v4du)__a | (__v4du)__b);
615 }
616 
617 /// Performs a bitwise OR of two 256-bit vectors of [8 x float].
618 ///
619 /// \headerfile <x86intrin.h>
620 ///
621 /// This intrinsic corresponds to the <c> VORPS </c> instruction.
622 ///
623 /// \param __a
624 ///    A 256-bit vector of [8 x float] containing one of the source operands.
625 /// \param __b
626 ///    A 256-bit vector of [8 x float] containing one of the source operands.
627 /// \returns A 256-bit vector of [8 x float] containing the bitwise OR of the
628 ///    values between both operands.
629 static __inline __m256 __DEFAULT_FN_ATTRS
630 _mm256_or_ps(__m256 __a, __m256 __b)
631 {
632   return (__m256)((__v8su)__a | (__v8su)__b);
633 }
634 
635 /// Performs a bitwise XOR of two 256-bit vectors of [4 x double].
636 ///
637 /// \headerfile <x86intrin.h>
638 ///
639 /// This intrinsic corresponds to the <c> VXORPD </c> instruction.
640 ///
641 /// \param __a
642 ///    A 256-bit vector of [4 x double] containing one of the source operands.
643 /// \param __b
644 ///    A 256-bit vector of [4 x double] containing one of the source operands.
645 /// \returns A 256-bit vector of [4 x double] containing the bitwise XOR of the
646 ///    values between both operands.
647 static __inline __m256d __DEFAULT_FN_ATTRS
648 _mm256_xor_pd(__m256d __a, __m256d __b)
649 {
650   return (__m256d)((__v4du)__a ^ (__v4du)__b);
651 }
652 
653 /// Performs a bitwise XOR of two 256-bit vectors of [8 x float].
654 ///
655 /// \headerfile <x86intrin.h>
656 ///
657 /// This intrinsic corresponds to the <c> VXORPS </c> instruction.
658 ///
659 /// \param __a
660 ///    A 256-bit vector of [8 x float] containing one of the source operands.
661 /// \param __b
662 ///    A 256-bit vector of [8 x float] containing one of the source operands.
663 /// \returns A 256-bit vector of [8 x float] containing the bitwise XOR of the
664 ///    values between both operands.
665 static __inline __m256 __DEFAULT_FN_ATTRS
666 _mm256_xor_ps(__m256 __a, __m256 __b)
667 {
668   return (__m256)((__v8su)__a ^ (__v8su)__b);
669 }
670 
671 /* Horizontal arithmetic */
672 /// Horizontally adds the adjacent pairs of values contained in two
673 ///    256-bit vectors of [4 x double].
674 ///
675 /// \headerfile <x86intrin.h>
676 ///
677 /// This intrinsic corresponds to the <c> VHADDPD </c> instruction.
678 ///
679 /// \param __a
680 ///    A 256-bit vector of [4 x double] containing one of the source operands.
681 ///    The horizontal sums of the values are returned in the even-indexed
682 ///    elements of a vector of [4 x double].
683 /// \param __b
684 ///    A 256-bit vector of [4 x double] containing one of the source operands.
685 ///    The horizontal sums of the values are returned in the odd-indexed
686 ///    elements of a vector of [4 x double].
687 /// \returns A 256-bit vector of [4 x double] containing the horizontal sums of
688 ///    both operands.
689 static __inline __m256d __DEFAULT_FN_ATTRS
690 _mm256_hadd_pd(__m256d __a, __m256d __b)
691 {
692   return (__m256d)__builtin_ia32_haddpd256((__v4df)__a, (__v4df)__b);
693 }
694 
695 /// Horizontally adds the adjacent pairs of values contained in two
696 ///    256-bit vectors of [8 x float].
697 ///
698 /// \headerfile <x86intrin.h>
699 ///
700 /// This intrinsic corresponds to the <c> VHADDPS </c> instruction.
701 ///
702 /// \param __a
703 ///    A 256-bit vector of [8 x float] containing one of the source operands.
704 ///    The horizontal sums of the values are returned in the elements with
705 ///    index 0, 1, 4, 5 of a vector of [8 x float].
706 /// \param __b
707 ///    A 256-bit vector of [8 x float] containing one of the source operands.
708 ///    The horizontal sums of the values are returned in the elements with
709 ///    index 2, 3, 6, 7 of a vector of [8 x float].
710 /// \returns A 256-bit vector of [8 x float] containing the horizontal sums of
711 ///    both operands.
712 static __inline __m256 __DEFAULT_FN_ATTRS
713 _mm256_hadd_ps(__m256 __a, __m256 __b)
714 {
715   return (__m256)__builtin_ia32_haddps256((__v8sf)__a, (__v8sf)__b);
716 }
717 
718 /// Horizontally subtracts the adjacent pairs of values contained in two
719 ///    256-bit vectors of [4 x double].
720 ///
721 /// \headerfile <x86intrin.h>
722 ///
723 /// This intrinsic corresponds to the <c> VHSUBPD </c> instruction.
724 ///
725 /// \param __a
726 ///    A 256-bit vector of [4 x double] containing one of the source operands.
727 ///    The horizontal differences between the values are returned in the
728 ///    even-indexed elements of a vector of [4 x double].
729 /// \param __b
730 ///    A 256-bit vector of [4 x double] containing one of the source operands.
731 ///    The horizontal differences between the values are returned in the
732 ///    odd-indexed elements of a vector of [4 x double].
733 /// \returns A 256-bit vector of [4 x double] containing the horizontal
734 ///    differences of both operands.
735 static __inline __m256d __DEFAULT_FN_ATTRS
736 _mm256_hsub_pd(__m256d __a, __m256d __b)
737 {
738   return (__m256d)__builtin_ia32_hsubpd256((__v4df)__a, (__v4df)__b);
739 }
740 
741 /// Horizontally subtracts the adjacent pairs of values contained in two
742 ///    256-bit vectors of [8 x float].
743 ///
744 /// \headerfile <x86intrin.h>
745 ///
746 /// This intrinsic corresponds to the <c> VHSUBPS </c> instruction.
747 ///
748 /// \param __a
749 ///    A 256-bit vector of [8 x float] containing one of the source operands.
750 ///    The horizontal differences between the values are returned in the
751 ///    elements with index 0, 1, 4, 5 of a vector of [8 x float].
752 /// \param __b
753 ///    A 256-bit vector of [8 x float] containing one of the source operands.
754 ///    The horizontal differences between the values are returned in the
755 ///    elements with index 2, 3, 6, 7 of a vector of [8 x float].
756 /// \returns A 256-bit vector of [8 x float] containing the horizontal
757 ///    differences of both operands.
758 static __inline __m256 __DEFAULT_FN_ATTRS
759 _mm256_hsub_ps(__m256 __a, __m256 __b)
760 {
761   return (__m256)__builtin_ia32_hsubps256((__v8sf)__a, (__v8sf)__b);
762 }
763 
764 /* Vector permutations */
765 /// Copies the values in a 128-bit vector of [2 x double] as specified
766 ///    by the 128-bit integer vector operand.
767 ///
768 /// \headerfile <x86intrin.h>
769 ///
770 /// This intrinsic corresponds to the <c> VPERMILPD </c> instruction.
771 ///
772 /// \param __a
773 ///    A 128-bit vector of [2 x double].
774 /// \param __c
775 ///    A 128-bit integer vector operand specifying how the values are to be
776 ///    copied. \n
777 ///    Bit [1]: \n
778 ///      0: Bits [63:0] of the source are copied to bits [63:0] of the returned
779 ///         vector. \n
780 ///      1: Bits [127:64] of the source are copied to bits [63:0] of the
781 ///         returned vector. \n
782 ///    Bit [65]: \n
783 ///      0: Bits [63:0] of the source are copied to bits [127:64] of the
784 ///         returned vector. \n
785 ///      1: Bits [127:64] of the source are copied to bits [127:64] of the
786 ///         returned vector.
787 /// \returns A 128-bit vector of [2 x double] containing the copied values.
788 static __inline __m128d __DEFAULT_FN_ATTRS128
789 _mm_permutevar_pd(__m128d __a, __m128i __c)
790 {
791   return (__m128d)__builtin_ia32_vpermilvarpd((__v2df)__a, (__v2di)__c);
792 }
793 
794 /// Copies the values in a 256-bit vector of [4 x double] as specified
795 ///    by the 256-bit integer vector operand.
796 ///
797 /// \headerfile <x86intrin.h>
798 ///
799 /// This intrinsic corresponds to the <c> VPERMILPD </c> instruction.
800 ///
801 /// \param __a
802 ///    A 256-bit vector of [4 x double].
803 /// \param __c
804 ///    A 256-bit integer vector operand specifying how the values are to be
805 ///    copied. \n
806 ///    Bit [1]: \n
807 ///      0: Bits [63:0] of the source are copied to bits [63:0] of the returned
808 ///         vector. \n
809 ///      1: Bits [127:64] of the source are copied to bits [63:0] of the
810 ///         returned vector. \n
811 ///    Bit [65]: \n
812 ///      0: Bits [63:0] of the source are copied to bits [127:64] of the
813 ///         returned vector. \n
814 ///      1: Bits [127:64] of the source are copied to bits [127:64] of the
815 ///         returned vector. \n
816 ///    Bit [129]: \n
817 ///      0: Bits [191:128] of the source are copied to bits [191:128] of the
818 ///         returned vector. \n
819 ///      1: Bits [255:192] of the source are copied to bits [191:128] of the
820 ///         returned vector. \n
821 ///    Bit [193]: \n
822 ///      0: Bits [191:128] of the source are copied to bits [255:192] of the
823 ///         returned vector. \n
824 ///      1: Bits [255:192] of the source are copied to bits [255:192] of the
825 ///    returned vector.
826 /// \returns A 256-bit vector of [4 x double] containing the copied values.
827 static __inline __m256d __DEFAULT_FN_ATTRS
828 _mm256_permutevar_pd(__m256d __a, __m256i __c)
829 {
830   return (__m256d)__builtin_ia32_vpermilvarpd256((__v4df)__a, (__v4di)__c);
831 }
832 
833 /// Copies the values stored in a 128-bit vector of [4 x float] as
834 ///    specified by the 128-bit integer vector operand.
835 /// \headerfile <x86intrin.h>
836 ///
837 /// This intrinsic corresponds to the <c> VPERMILPS </c> instruction.
838 ///
839 /// \param __a
840 ///    A 128-bit vector of [4 x float].
841 /// \param __c
842 ///    A 128-bit integer vector operand specifying how the values are to be
843 ///    copied. \n
844 ///    Bits [1:0]: \n
845 ///      00: Bits [31:0] of the source are copied to bits [31:0] of the
846 ///          returned vector. \n
847 ///      01: Bits [63:32] of the source are copied to bits [31:0] of the
848 ///          returned vector. \n
849 ///      10: Bits [95:64] of the source are copied to bits [31:0] of the
850 ///          returned vector. \n
851 ///      11: Bits [127:96] of the source are copied to bits [31:0] of the
852 ///          returned vector. \n
853 ///    Bits [33:32]: \n
854 ///      00: Bits [31:0] of the source are copied to bits [63:32] of the
855 ///          returned vector. \n
856 ///      01: Bits [63:32] of the source are copied to bits [63:32] of the
857 ///          returned vector. \n
858 ///      10: Bits [95:64] of the source are copied to bits [63:32] of the
859 ///          returned vector. \n
860 ///      11: Bits [127:96] of the source are copied to bits [63:32] of the
861 ///          returned vector. \n
862 ///    Bits [65:64]: \n
863 ///      00: Bits [31:0] of the source are copied to bits [95:64] of the
864 ///          returned vector. \n
865 ///      01: Bits [63:32] of the source are copied to bits [95:64] of the
866 ///          returned vector. \n
867 ///      10: Bits [95:64] of the source are copied to bits [95:64] of the
868 ///          returned vector. \n
869 ///      11: Bits [127:96] of the source are copied to bits [95:64] of the
870 ///          returned vector. \n
871 ///    Bits [97:96]: \n
872 ///      00: Bits [31:0] of the source are copied to bits [127:96] of the
873 ///          returned vector. \n
874 ///      01: Bits [63:32] of the source are copied to bits [127:96] of the
875 ///          returned vector. \n
876 ///      10: Bits [95:64] of the source are copied to bits [127:96] of the
877 ///          returned vector. \n
878 ///      11: Bits [127:96] of the source are copied to bits [127:96] of the
879 ///          returned vector.
880 /// \returns A 128-bit vector of [4 x float] containing the copied values.
881 static __inline __m128 __DEFAULT_FN_ATTRS128
882 _mm_permutevar_ps(__m128 __a, __m128i __c)
883 {
884   return (__m128)__builtin_ia32_vpermilvarps((__v4sf)__a, (__v4si)__c);
885 }
886 
887 /// Copies the values stored in a 256-bit vector of [8 x float] as
888 ///    specified by the 256-bit integer vector operand.
889 ///
890 /// \headerfile <x86intrin.h>
891 ///
892 /// This intrinsic corresponds to the <c> VPERMILPS </c> instruction.
893 ///
894 /// \param __a
895 ///    A 256-bit vector of [8 x float].
896 /// \param __c
897 ///    A 256-bit integer vector operand specifying how the values are to be
898 ///    copied. \n
899 ///    Bits [1:0]: \n
900 ///      00: Bits [31:0] of the source are copied to bits [31:0] of the
901 ///          returned vector. \n
902 ///      01: Bits [63:32] of the source are copied to bits [31:0] of the
903 ///          returned vector. \n
904 ///      10: Bits [95:64] of the source are copied to bits [31:0] of the
905 ///          returned vector. \n
906 ///      11: Bits [127:96] of the source are copied to bits [31:0] of the
907 ///          returned vector. \n
908 ///    Bits [33:32]: \n
909 ///      00: Bits [31:0] of the source are copied to bits [63:32] of the
910 ///          returned vector. \n
911 ///      01: Bits [63:32] of the source are copied to bits [63:32] of the
912 ///          returned vector. \n
913 ///      10: Bits [95:64] of the source are copied to bits [63:32] of the
914 ///          returned vector. \n
915 ///      11: Bits [127:96] of the source are copied to bits [63:32] of the
916 ///          returned vector. \n
917 ///    Bits [65:64]: \n
918 ///      00: Bits [31:0] of the source are copied to bits [95:64] of the
919 ///          returned vector. \n
920 ///      01: Bits [63:32] of the source are copied to bits [95:64] of the
921 ///          returned vector. \n
922 ///      10: Bits [95:64] of the source are copied to bits [95:64] of the
923 ///          returned vector. \n
924 ///      11: Bits [127:96] of the source are copied to bits [95:64] of the
925 ///          returned vector. \n
926 ///    Bits [97:96]: \n
927 ///      00: Bits [31:0] of the source are copied to bits [127:96] of the
928 ///          returned vector. \n
929 ///      01: Bits [63:32] of the source are copied to bits [127:96] of the
930 ///          returned vector. \n
931 ///      10: Bits [95:64] of the source are copied to bits [127:96] of the
932 ///          returned vector. \n
933 ///      11: Bits [127:96] of the source are copied to bits [127:96] of the
934 ///          returned vector. \n
935 ///    Bits [129:128]: \n
936 ///      00: Bits [159:128] of the source are copied to bits [159:128] of the
937 ///          returned vector. \n
938 ///      01: Bits [191:160] of the source are copied to bits [159:128] of the
939 ///          returned vector. \n
940 ///      10: Bits [223:192] of the source are copied to bits [159:128] of the
941 ///          returned vector. \n
942 ///      11: Bits [255:224] of the source are copied to bits [159:128] of the
943 ///          returned vector. \n
944 ///    Bits [161:160]: \n
945 ///      00: Bits [159:128] of the source are copied to bits [191:160] of the
946 ///          returned vector. \n
947 ///      01: Bits [191:160] of the source are copied to bits [191:160] of the
948 ///          returned vector. \n
949 ///      10: Bits [223:192] of the source are copied to bits [191:160] of the
950 ///          returned vector. \n
951 ///      11: Bits [255:224] of the source are copied to bits [191:160] of the
952 ///          returned vector. \n
953 ///    Bits [193:192]: \n
954 ///      00: Bits [159:128] of the source are copied to bits [223:192] of the
955 ///          returned vector. \n
956 ///      01: Bits [191:160] of the source are copied to bits [223:192] of the
957 ///          returned vector. \n
958 ///      10: Bits [223:192] of the source are copied to bits [223:192] of the
959 ///          returned vector. \n
960 ///      11: Bits [255:224] of the source are copied to bits [223:192] of the
961 ///          returned vector. \n
962 ///    Bits [225:224]: \n
963 ///      00: Bits [159:128] of the source are copied to bits [255:224] of the
964 ///          returned vector. \n
965 ///      01: Bits [191:160] of the source are copied to bits [255:224] of the
966 ///          returned vector. \n
967 ///      10: Bits [223:192] of the source are copied to bits [255:224] of the
968 ///          returned vector. \n
969 ///      11: Bits [255:224] of the source are copied to bits [255:224] of the
970 ///          returned vector.
971 /// \returns A 256-bit vector of [8 x float] containing the copied values.
972 static __inline __m256 __DEFAULT_FN_ATTRS
973 _mm256_permutevar_ps(__m256 __a, __m256i __c)
974 {
975   return (__m256)__builtin_ia32_vpermilvarps256((__v8sf)__a, (__v8si)__c);
976 }
977 
978 /// Copies the values in a 128-bit vector of [2 x double] as specified
979 ///    by the immediate integer operand.
980 ///
981 /// \headerfile <x86intrin.h>
982 ///
983 /// \code
984 /// __m128d _mm_permute_pd(__m128d A, const int C);
985 /// \endcode
986 ///
987 /// This intrinsic corresponds to the <c> VPERMILPD </c> instruction.
988 ///
989 /// \param A
990 ///    A 128-bit vector of [2 x double].
991 /// \param C
992 ///    An immediate integer operand specifying how the values are to be
993 ///    copied. \n
994 ///    Bit [0]: \n
995 ///      0: Bits [63:0] of the source are copied to bits [63:0] of the returned
996 ///         vector. \n
997 ///      1: Bits [127:64] of the source are copied to bits [63:0] of the
998 ///         returned vector. \n
999 ///    Bit [1]: \n
1000 ///      0: Bits [63:0] of the source are copied to bits [127:64] of the
1001 ///         returned vector. \n
1002 ///      1: Bits [127:64] of the source are copied to bits [127:64] of the
1003 ///         returned vector.
1004 /// \returns A 128-bit vector of [2 x double] containing the copied values.
1005 #define _mm_permute_pd(A, C) \
1006   ((__m128d)__builtin_ia32_vpermilpd((__v2df)(__m128d)(A), (int)(C)))
1007 
1008 /// Copies the values in a 256-bit vector of [4 x double] as specified by
1009 ///    the immediate integer operand.
1010 ///
1011 /// \headerfile <x86intrin.h>
1012 ///
1013 /// \code
1014 /// __m256d _mm256_permute_pd(__m256d A, const int C);
1015 /// \endcode
1016 ///
1017 /// This intrinsic corresponds to the <c> VPERMILPD </c> instruction.
1018 ///
1019 /// \param A
1020 ///    A 256-bit vector of [4 x double].
1021 /// \param C
1022 ///    An immediate integer operand specifying how the values are to be
1023 ///    copied. \n
1024 ///    Bit [0]: \n
1025 ///      0: Bits [63:0] of the source are copied to bits [63:0] of the returned
1026 ///         vector. \n
1027 ///      1: Bits [127:64] of the source are copied to bits [63:0] of the
1028 ///         returned vector. \n
1029 ///    Bit [1]: \n
1030 ///      0: Bits [63:0] of the source are copied to bits [127:64] of the
1031 ///         returned vector. \n
1032 ///      1: Bits [127:64] of the source are copied to bits [127:64] of the
1033 ///         returned vector. \n
1034 ///    Bit [2]: \n
1035 ///      0: Bits [191:128] of the source are copied to bits [191:128] of the
1036 ///         returned vector. \n
1037 ///      1: Bits [255:192] of the source are copied to bits [191:128] of the
1038 ///         returned vector. \n
1039 ///    Bit [3]: \n
1040 ///      0: Bits [191:128] of the source are copied to bits [255:192] of the
1041 ///         returned vector. \n
1042 ///      1: Bits [255:192] of the source are copied to bits [255:192] of the
1043 ///         returned vector.
1044 /// \returns A 256-bit vector of [4 x double] containing the copied values.
1045 #define _mm256_permute_pd(A, C) \
1046   ((__m256d)__builtin_ia32_vpermilpd256((__v4df)(__m256d)(A), (int)(C)))
1047 
1048 /// Copies the values in a 128-bit vector of [4 x float] as specified by
1049 ///    the immediate integer operand.
1050 ///
1051 /// \headerfile <x86intrin.h>
1052 ///
1053 /// \code
1054 /// __m128 _mm_permute_ps(__m128 A, const int C);
1055 /// \endcode
1056 ///
1057 /// This intrinsic corresponds to the <c> VPERMILPS </c> instruction.
1058 ///
1059 /// \param A
1060 ///    A 128-bit vector of [4 x float].
1061 /// \param C
1062 ///    An immediate integer operand specifying how the values are to be
1063 ///    copied. \n
1064 ///    Bits [1:0]: \n
1065 ///      00: Bits [31:0] of the source are copied to bits [31:0] of the
1066 ///          returned vector. \n
1067 ///      01: Bits [63:32] of the source are copied to bits [31:0] of the
1068 ///          returned vector. \n
1069 ///      10: Bits [95:64] of the source are copied to bits [31:0] of the
1070 ///          returned vector. \n
1071 ///      11: Bits [127:96] of the source are copied to bits [31:0] of the
1072 ///          returned vector. \n
1073 ///    Bits [3:2]: \n
1074 ///      00: Bits [31:0] of the source are copied to bits [63:32] of the
1075 ///          returned vector. \n
1076 ///      01: Bits [63:32] of the source are copied to bits [63:32] of the
1077 ///          returned vector. \n
1078 ///      10: Bits [95:64] of the source are copied to bits [63:32] of the
1079 ///          returned vector. \n
1080 ///      11: Bits [127:96] of the source are copied to bits [63:32] of the
1081 ///          returned vector. \n
1082 ///    Bits [5:4]: \n
1083 ///      00: Bits [31:0] of the source are copied to bits [95:64] of the
1084 ///          returned vector. \n
1085 ///      01: Bits [63:32] of the source are copied to bits [95:64] of the
1086 ///          returned vector. \n
1087 ///      10: Bits [95:64] of the source are copied to bits [95:64] of the
1088 ///          returned vector. \n
1089 ///      11: Bits [127:96] of the source are copied to bits [95:64] of the
1090 ///          returned vector. \n
1091 ///    Bits [7:6]: \n
1092 ///      00: Bits [31:0] of the source are copied to bits [127:96] of the
1093 ///          returned vector. \n
1094 ///      01: Bits [63:32] of the source are copied to bits [127:96] of the
1095 ///          returned vector. \n
1096 ///      10: Bits [95:64] of the source are copied to bits [127:96] of the
1097 ///          returned vector. \n
1098 ///      11: Bits [127:96] of the source are copied to bits [127:96] of the
1099 ///          returned vector.
1100 /// \returns A 128-bit vector of [4 x float] containing the copied values.
1101 #define _mm_permute_ps(A, C) \
1102   ((__m128)__builtin_ia32_vpermilps((__v4sf)(__m128)(A), (int)(C)))
1103 
1104 /// Copies the values in a 256-bit vector of [8 x float] as specified by
1105 ///    the immediate integer operand.
1106 ///
1107 /// \headerfile <x86intrin.h>
1108 ///
1109 /// \code
1110 /// __m256 _mm256_permute_ps(__m256 A, const int C);
1111 /// \endcode
1112 ///
1113 /// This intrinsic corresponds to the <c> VPERMILPS </c> instruction.
1114 ///
1115 /// \param A
1116 ///    A 256-bit vector of [8 x float].
1117 /// \param C
1118 ///    An immediate integer operand specifying how the values are to be
1119 ///    copied. \n
1120 ///    Bits [1:0]: \n
1121 ///      00: Bits [31:0] of the source are copied to bits [31:0] of the
1122 ///          returned vector. \n
1123 ///      01: Bits [63:32] of the source are copied to bits [31:0] of the
1124 ///          returned vector. \n
1125 ///      10: Bits [95:64] of the source are copied to bits [31:0] of the
1126 ///          returned vector. \n
1127 ///      11: Bits [127:96] of the source are copied to bits [31:0] of the
1128 ///          returned vector. \n
1129 ///    Bits [3:2]: \n
1130 ///      00: Bits [31:0] of the source are copied to bits [63:32] of the
1131 ///          returned vector. \n
1132 ///      01: Bits [63:32] of the source are copied to bits [63:32] of the
1133 ///          returned vector. \n
1134 ///      10: Bits [95:64] of the source are copied to bits [63:32] of the
1135 ///          returned vector. \n
1136 ///      11: Bits [127:96] of the source are copied to bits [63:32] of the
1137 ///          returned vector. \n
1138 ///    Bits [5:4]: \n
1139 ///      00: Bits [31:0] of the source are copied to bits [95:64] of the
1140 ///          returned vector. \n
1141 ///      01: Bits [63:32] of the source are copied to bits [95:64] of the
1142 ///          returned vector. \n
1143 ///      10: Bits [95:64] of the source are copied to bits [95:64] of the
1144 ///          returned vector. \n
1145 ///      11: Bits [127:96] of the source are copied to bits [95:64] of the
1146 ///          returned vector. \n
1147 ///    Bits [7:6]: \n
1148 ///      00: Bits [31:0] of the source are copied to bits [127:96] of the
1149 ///          returned vector. \n
1150 ///      01: Bits [63:32] of the source are copied to bits [127:96] of the
1151 ///          returned vector. \n
1152 ///      10: Bits [95:64] of the source are copied to bits [127:96] of the
1153 ///          returned vector. \n
1154 ///      11: Bits [127:96] of the source are copied to bits [127:96] of the
1155 ///          returned vector. \n
1156 ///    Bits [1:0]: \n
1157 ///      00: Bits [159:128] of the source are copied to bits [159:128] of the
1158 ///          returned vector. \n
1159 ///      01: Bits [191:160] of the source are copied to bits [159:128] of the
1160 ///          returned vector. \n
1161 ///      10: Bits [223:192] of the source are copied to bits [159:128] of the
1162 ///          returned vector. \n
1163 ///      11: Bits [255:224] of the source are copied to bits [159:128] of the
1164 ///          returned vector. \n
1165 ///    Bits [3:2]: \n
1166 ///      00: Bits [159:128] of the source are copied to bits [191:160] of the
1167 ///          returned vector. \n
1168 ///      01: Bits [191:160] of the source are copied to bits [191:160] of the
1169 ///          returned vector. \n
1170 ///      10: Bits [223:192] of the source are copied to bits [191:160] of the
1171 ///          returned vector. \n
1172 ///      11: Bits [255:224] of the source are copied to bits [191:160] of the
1173 ///          returned vector. \n
1174 ///    Bits [5:4]: \n
1175 ///      00: Bits [159:128] of the source are copied to bits [223:192] of the
1176 ///          returned vector. \n
1177 ///      01: Bits [191:160] of the source are copied to bits [223:192] of the
1178 ///          returned vector. \n
1179 ///      10: Bits [223:192] of the source are copied to bits [223:192] of the
1180 ///          returned vector. \n
1181 ///      11: Bits [255:224] of the source are copied to bits [223:192] of the
1182 ///          returned vector. \n
1183 ///    Bits [7:6]: \n
1184 ///      00: Bits [159:128] of the source are copied to bits [255:224] of the
1185 ///          returned vector. \n
1186 ///      01: Bits [191:160] of the source are copied to bits [255:224] of the
1187 ///          returned vector. \n
1188 ///      10: Bits [223:192] of the source are copied to bits [255:224] of the
1189 ///          returned vector. \n
1190 ///      11: Bits [255:224] of the source are copied to bits [255:224] of the
1191 ///          returned vector.
1192 /// \returns A 256-bit vector of [8 x float] containing the copied values.
1193 #define _mm256_permute_ps(A, C) \
1194   ((__m256)__builtin_ia32_vpermilps256((__v8sf)(__m256)(A), (int)(C)))
1195 
1196 /// Permutes 128-bit data values stored in two 256-bit vectors of
1197 ///    [4 x double], as specified by the immediate integer operand.
1198 ///
1199 /// \headerfile <x86intrin.h>
1200 ///
1201 /// \code
1202 /// __m256d _mm256_permute2f128_pd(__m256d V1, __m256d V2, const int M);
1203 /// \endcode
1204 ///
1205 /// This intrinsic corresponds to the <c> VPERM2F128 </c> instruction.
1206 ///
1207 /// \param V1
1208 ///    A 256-bit vector of [4 x double].
1209 /// \param V2
1210 ///    A 256-bit vector of [4 x double.
1211 /// \param M
1212 ///    An immediate integer operand specifying how the values are to be
1213 ///    permuted. \n
1214 ///    Bits [1:0]: \n
1215 ///      00: Bits [127:0] of operand \a V1 are copied to bits [127:0] of the
1216 ///          destination. \n
1217 ///      01: Bits [255:128] of operand \a V1 are copied to bits [127:0] of the
1218 ///          destination. \n
1219 ///      10: Bits [127:0] of operand \a V2 are copied to bits [127:0] of the
1220 ///          destination. \n
1221 ///      11: Bits [255:128] of operand \a V2 are copied to bits [127:0] of the
1222 ///          destination. \n
1223 ///    Bits [5:4]: \n
1224 ///      00: Bits [127:0] of operand \a V1 are copied to bits [255:128] of the
1225 ///          destination. \n
1226 ///      01: Bits [255:128] of operand \a V1 are copied to bits [255:128] of the
1227 ///          destination. \n
1228 ///      10: Bits [127:0] of operand \a V2 are copied to bits [255:128] of the
1229 ///          destination. \n
1230 ///      11: Bits [255:128] of operand \a V2 are copied to bits [255:128] of the
1231 ///          destination.
1232 /// \returns A 256-bit vector of [4 x double] containing the copied values.
1233 #define _mm256_permute2f128_pd(V1, V2, M) \
1234   ((__m256d)__builtin_ia32_vperm2f128_pd256((__v4df)(__m256d)(V1), \
1235                                             (__v4df)(__m256d)(V2), (int)(M)))
1236 
1237 /// Permutes 128-bit data values stored in two 256-bit vectors of
1238 ///    [8 x float], as specified by the immediate integer operand.
1239 ///
1240 /// \headerfile <x86intrin.h>
1241 ///
1242 /// \code
1243 /// __m256 _mm256_permute2f128_ps(__m256 V1, __m256 V2, const int M);
1244 /// \endcode
1245 ///
1246 /// This intrinsic corresponds to the <c> VPERM2F128 </c> instruction.
1247 ///
1248 /// \param V1
1249 ///    A 256-bit vector of [8 x float].
1250 /// \param V2
1251 ///    A 256-bit vector of [8 x float].
1252 /// \param M
1253 ///    An immediate integer operand specifying how the values are to be
1254 ///    permuted. \n
1255 ///    Bits [1:0]: \n
1256 ///    00: Bits [127:0] of operand \a V1 are copied to bits [127:0] of the
1257 ///    destination. \n
1258 ///    01: Bits [255:128] of operand \a V1 are copied to bits [127:0] of the
1259 ///    destination. \n
1260 ///    10: Bits [127:0] of operand \a V2 are copied to bits [127:0] of the
1261 ///    destination. \n
1262 ///    11: Bits [255:128] of operand \a V2 are copied to bits [127:0] of the
1263 ///    destination. \n
1264 ///    Bits [5:4]: \n
1265 ///    00: Bits [127:0] of operand \a V1 are copied to bits [255:128] of the
1266 ///    destination. \n
1267 ///    01: Bits [255:128] of operand \a V1 are copied to bits [255:128] of the
1268 ///    destination. \n
1269 ///    10: Bits [127:0] of operand \a V2 are copied to bits [255:128] of the
1270 ///    destination. \n
1271 ///    11: Bits [255:128] of operand \a V2 are copied to bits [255:128] of the
1272 ///    destination.
1273 /// \returns A 256-bit vector of [8 x float] containing the copied values.
1274 #define _mm256_permute2f128_ps(V1, V2, M) \
1275   ((__m256)__builtin_ia32_vperm2f128_ps256((__v8sf)(__m256)(V1), \
1276                                            (__v8sf)(__m256)(V2), (int)(M)))
1277 
1278 /// Permutes 128-bit data values stored in two 256-bit integer vectors,
1279 ///    as specified by the immediate integer operand.
1280 ///
1281 /// \headerfile <x86intrin.h>
1282 ///
1283 /// \code
1284 /// __m256i _mm256_permute2f128_si256(__m256i V1, __m256i V2, const int M);
1285 /// \endcode
1286 ///
1287 /// This intrinsic corresponds to the <c> VPERM2F128 </c> instruction.
1288 ///
1289 /// \param V1
1290 ///    A 256-bit integer vector.
1291 /// \param V2
1292 ///    A 256-bit integer vector.
1293 /// \param M
1294 ///    An immediate integer operand specifying how the values are to be copied.
1295 ///    Bits [1:0]: \n
1296 ///    00: Bits [127:0] of operand \a V1 are copied to bits [127:0] of the
1297 ///    destination. \n
1298 ///    01: Bits [255:128] of operand \a V1 are copied to bits [127:0] of the
1299 ///    destination. \n
1300 ///    10: Bits [127:0] of operand \a V2 are copied to bits [127:0] of the
1301 ///    destination. \n
1302 ///    11: Bits [255:128] of operand \a V2 are copied to bits [127:0] of the
1303 ///    destination. \n
1304 ///    Bits [5:4]: \n
1305 ///    00: Bits [127:0] of operand \a V1 are copied to bits [255:128] of the
1306 ///    destination. \n
1307 ///    01: Bits [255:128] of operand \a V1 are copied to bits [255:128] of the
1308 ///    destination. \n
1309 ///    10: Bits [127:0] of operand \a V2 are copied to bits [255:128] of the
1310 ///    destination. \n
1311 ///    11: Bits [255:128] of operand \a V2 are copied to bits [255:128] of the
1312 ///    destination.
1313 /// \returns A 256-bit integer vector containing the copied values.
1314 #define _mm256_permute2f128_si256(V1, V2, M) \
1315   ((__m256i)__builtin_ia32_vperm2f128_si256((__v8si)(__m256i)(V1), \
1316                                             (__v8si)(__m256i)(V2), (int)(M)))
1317 
1318 /* Vector Blend */
1319 /// Merges 64-bit double-precision data values stored in either of the
1320 ///    two 256-bit vectors of [4 x double], as specified by the immediate
1321 ///    integer operand.
1322 ///
1323 /// \headerfile <x86intrin.h>
1324 ///
1325 /// \code
1326 /// __m256d _mm256_blend_pd(__m256d V1, __m256d V2, const int M);
1327 /// \endcode
1328 ///
1329 /// This intrinsic corresponds to the <c> VBLENDPD </c> instruction.
1330 ///
1331 /// \param V1
1332 ///    A 256-bit vector of [4 x double].
1333 /// \param V2
1334 ///    A 256-bit vector of [4 x double].
1335 /// \param M
1336 ///    An immediate integer operand, with mask bits [3:0] specifying how the
1337 ///    values are to be copied. The position of the mask bit corresponds to the
1338 ///    index of a copied value. When a mask bit is 0, the corresponding 64-bit
1339 ///    element in operand \a V1 is copied to the same position in the
1340 ///    destination. When a mask bit is 1, the corresponding 64-bit element in
1341 ///    operand \a V2 is copied to the same position in the destination.
1342 /// \returns A 256-bit vector of [4 x double] containing the copied values.
1343 #define _mm256_blend_pd(V1, V2, M) \
1344   ((__m256d)__builtin_ia32_blendpd256((__v4df)(__m256d)(V1), \
1345                                       (__v4df)(__m256d)(V2), (int)(M)))
1346 
1347 /// Merges 32-bit single-precision data values stored in either of the
1348 ///    two 256-bit vectors of [8 x float], as specified by the immediate
1349 ///    integer operand.
1350 ///
1351 /// \headerfile <x86intrin.h>
1352 ///
1353 /// \code
1354 /// __m256 _mm256_blend_ps(__m256 V1, __m256 V2, const int M);
1355 /// \endcode
1356 ///
1357 /// This intrinsic corresponds to the <c> VBLENDPS </c> instruction.
1358 ///
1359 /// \param V1
1360 ///    A 256-bit vector of [8 x float].
1361 /// \param V2
1362 ///    A 256-bit vector of [8 x float].
1363 /// \param M
1364 ///    An immediate integer operand, with mask bits [7:0] specifying how the
1365 ///    values are to be copied. The position of the mask bit corresponds to the
1366 ///    index of a copied value. When a mask bit is 0, the corresponding 32-bit
1367 ///    element in operand \a V1 is copied to the same position in the
1368 ///    destination. When a mask bit is 1, the corresponding 32-bit element in
1369 ///    operand \a V2 is copied to the same position in the destination.
1370 /// \returns A 256-bit vector of [8 x float] containing the copied values.
1371 #define _mm256_blend_ps(V1, V2, M) \
1372   ((__m256)__builtin_ia32_blendps256((__v8sf)(__m256)(V1), \
1373                                      (__v8sf)(__m256)(V2), (int)(M)))
1374 
1375 /// Merges 64-bit double-precision data values stored in either of the
1376 ///    two 256-bit vectors of [4 x double], as specified by the 256-bit vector
1377 ///    operand.
1378 ///
1379 /// \headerfile <x86intrin.h>
1380 ///
1381 /// This intrinsic corresponds to the <c> VBLENDVPD </c> instruction.
1382 ///
1383 /// \param __a
1384 ///    A 256-bit vector of [4 x double].
1385 /// \param __b
1386 ///    A 256-bit vector of [4 x double].
1387 /// \param __c
1388 ///    A 256-bit vector operand, with mask bits 255, 191, 127, and 63 specifying
1389 ///    how the values are to be copied. The position of the mask bit corresponds
1390 ///    to the most significant bit of a copied value. When a mask bit is 0, the
1391 ///    corresponding 64-bit element in operand \a __a is copied to the same
1392 ///    position in the destination. When a mask bit is 1, the corresponding
1393 ///    64-bit element in operand \a __b is copied to the same position in the
1394 ///    destination.
1395 /// \returns A 256-bit vector of [4 x double] containing the copied values.
1396 static __inline __m256d __DEFAULT_FN_ATTRS
1397 _mm256_blendv_pd(__m256d __a, __m256d __b, __m256d __c)
1398 {
1399   return (__m256d)__builtin_ia32_blendvpd256(
1400     (__v4df)__a, (__v4df)__b, (__v4df)__c);
1401 }
1402 
1403 /// Merges 32-bit single-precision data values stored in either of the
1404 ///    two 256-bit vectors of [8 x float], as specified by the 256-bit vector
1405 ///    operand.
1406 ///
1407 /// \headerfile <x86intrin.h>
1408 ///
1409 /// This intrinsic corresponds to the <c> VBLENDVPS </c> instruction.
1410 ///
1411 /// \param __a
1412 ///    A 256-bit vector of [8 x float].
1413 /// \param __b
1414 ///    A 256-bit vector of [8 x float].
1415 /// \param __c
1416 ///    A 256-bit vector operand, with mask bits 255, 223, 191, 159, 127, 95, 63,
1417 ///    and 31 specifying how the values are to be copied. The position of the
1418 ///    mask bit corresponds to the most significant bit of a copied value. When
1419 ///    a mask bit is 0, the corresponding 32-bit element in operand \a __a is
1420 ///    copied to the same position in the destination. When a mask bit is 1, the
1421 ///    corresponding 32-bit element in operand \a __b is copied to the same
1422 ///    position in the destination.
1423 /// \returns A 256-bit vector of [8 x float] containing the copied values.
1424 static __inline __m256 __DEFAULT_FN_ATTRS
1425 _mm256_blendv_ps(__m256 __a, __m256 __b, __m256 __c)
1426 {
1427   return (__m256)__builtin_ia32_blendvps256(
1428     (__v8sf)__a, (__v8sf)__b, (__v8sf)__c);
1429 }
1430 
1431 /* Vector Dot Product */
1432 /// Computes two dot products in parallel, using the lower and upper
1433 ///    halves of two [8 x float] vectors as input to the two computations, and
1434 ///    returning the two dot products in the lower and upper halves of the
1435 ///    [8 x float] result.
1436 ///
1437 ///    The immediate integer operand controls which input elements will
1438 ///    contribute to the dot product, and where the final results are returned.
1439 ///    In general, for each dot product, the four corresponding elements of the
1440 ///    input vectors are multiplied; the first two and second two products are
1441 ///    summed, then the two sums are added to form the final result.
1442 ///
1443 /// \headerfile <x86intrin.h>
1444 ///
1445 /// \code
1446 /// __m256 _mm256_dp_ps(__m256 V1, __m256 V2, const int M);
1447 /// \endcode
1448 ///
1449 /// This intrinsic corresponds to the <c> VDPPS </c> instruction.
1450 ///
1451 /// \param V1
1452 ///    A vector of [8 x float] values, treated as two [4 x float] vectors.
1453 /// \param V2
1454 ///    A vector of [8 x float] values, treated as two [4 x float] vectors.
1455 /// \param M
1456 ///    An immediate integer argument. Bits [7:4] determine which elements of
1457 ///    the input vectors are used, with bit [4] corresponding to the lowest
1458 ///    element and bit [7] corresponding to the highest element of each [4 x
1459 ///    float] subvector. If a bit is set, the corresponding elements from the
1460 ///    two input vectors are used as an input for dot product; otherwise that
1461 ///    input is treated as zero. Bits [3:0] determine which elements of the
1462 ///    result will receive a copy of the final dot product, with bit [0]
1463 ///    corresponding to the lowest element and bit [3] corresponding to the
1464 ///    highest element of each [4 x float] subvector. If a bit is set, the dot
1465 ///    product is returned in the corresponding element; otherwise that element
1466 ///    is set to zero. The bitmask is applied in the same way to each of the
1467 ///    two parallel dot product computations.
1468 /// \returns A 256-bit vector of [8 x float] containing the two dot products.
1469 #define _mm256_dp_ps(V1, V2, M) \
1470   ((__m256)__builtin_ia32_dpps256((__v8sf)(__m256)(V1), \
1471                                   (__v8sf)(__m256)(V2), (M)))
1472 
1473 /* Vector shuffle */
1474 /// Selects 8 float values from the 256-bit operands of [8 x float], as
1475 ///    specified by the immediate value operand.
1476 ///
1477 ///    The four selected elements in each operand are copied to the destination
1478 ///    according to the bits specified in the immediate operand. The selected
1479 ///    elements from the first 256-bit operand are copied to bits [63:0] and
1480 ///    bits [191:128] of the destination, and the selected elements from the
1481 ///    second 256-bit operand are copied to bits [127:64] and bits [255:192] of
1482 ///    the destination. For example, if bits [7:0] of the immediate operand
1483 ///    contain a value of 0xFF, the 256-bit destination vector would contain the
1484 ///    following values: b[7], b[7], a[7], a[7], b[3], b[3], a[3], a[3].
1485 ///
1486 /// \headerfile <x86intrin.h>
1487 ///
1488 /// \code
1489 /// __m256 _mm256_shuffle_ps(__m256 a, __m256 b, const int mask);
1490 /// \endcode
1491 ///
1492 /// This intrinsic corresponds to the <c> VSHUFPS </c> instruction.
1493 ///
1494 /// \param a
1495 ///    A 256-bit vector of [8 x float]. The four selected elements in this
1496 ///    operand are copied to bits [63:0] and bits [191:128] in the destination,
1497 ///    according to the bits specified in the immediate operand.
1498 /// \param b
1499 ///    A 256-bit vector of [8 x float]. The four selected elements in this
1500 ///    operand are copied to bits [127:64] and bits [255:192] in the
1501 ///    destination, according to the bits specified in the immediate operand.
1502 /// \param mask
1503 ///    An immediate value containing an 8-bit value specifying which elements to
1504 ///    copy from \a a and \a b \n.
1505 ///    Bits [3:0] specify the values copied from operand \a a. \n
1506 ///    Bits [7:4] specify the values copied from operand \a b. \n
1507 ///    The destinations within the 256-bit destination are assigned values as
1508 ///    follows, according to the bit value assignments described below: \n
1509 ///    Bits [1:0] are used to assign values to bits [31:0] and [159:128] in the
1510 ///    destination. \n
1511 ///    Bits [3:2] are used to assign values to bits [63:32] and [191:160] in the
1512 ///    destination. \n
1513 ///    Bits [5:4] are used to assign values to bits [95:64] and [223:192] in the
1514 ///    destination. \n
1515 ///    Bits [7:6] are used to assign values to bits [127:96] and [255:224] in
1516 ///    the destination. \n
1517 ///    Bit value assignments: \n
1518 ///    00: Bits [31:0] and [159:128] are copied from the selected operand. \n
1519 ///    01: Bits [63:32] and [191:160] are copied from the selected operand. \n
1520 ///    10: Bits [95:64] and [223:192] are copied from the selected operand. \n
1521 ///    11: Bits [127:96] and [255:224] are copied from the selected operand. \n
1522 ///    Note: To generate a mask, you can use the \c _MM_SHUFFLE macro.
1523 ///    <c>_MM_SHUFFLE(b6, b4, b2, b0)</c> can create an 8-bit mask of the form
1524 ///    <c>[b6, b4, b2, b0]</c>.
1525 /// \returns A 256-bit vector of [8 x float] containing the shuffled values.
1526 #define _mm256_shuffle_ps(a, b, mask) \
1527   ((__m256)__builtin_ia32_shufps256((__v8sf)(__m256)(a), \
1528                                     (__v8sf)(__m256)(b), (int)(mask)))
1529 
1530 /// Selects four double-precision values from the 256-bit operands of
1531 ///    [4 x double], as specified by the immediate value operand.
1532 ///
1533 ///    The selected elements from the first 256-bit operand are copied to bits
1534 ///    [63:0] and bits [191:128] in the destination, and the selected elements
1535 ///    from the second 256-bit operand are copied to bits [127:64] and bits
1536 ///    [255:192] in the destination. For example, if bits [3:0] of the immediate
1537 ///    operand contain a value of 0xF, the 256-bit destination vector would
1538 ///    contain the following values: b[3], a[3], b[1], a[1].
1539 ///
1540 /// \headerfile <x86intrin.h>
1541 ///
1542 /// \code
1543 /// __m256d _mm256_shuffle_pd(__m256d a, __m256d b, const int mask);
1544 /// \endcode
1545 ///
1546 /// This intrinsic corresponds to the <c> VSHUFPD </c> instruction.
1547 ///
1548 /// \param a
1549 ///    A 256-bit vector of [4 x double].
1550 /// \param b
1551 ///    A 256-bit vector of [4 x double].
1552 /// \param mask
1553 ///    An immediate value containing 8-bit values specifying which elements to
1554 ///    copy from \a a and \a b: \n
1555 ///    Bit [0]=0: Bits [63:0] are copied from \a a to bits [63:0] of the
1556 ///    destination. \n
1557 ///    Bit [0]=1: Bits [127:64] are copied from \a a to bits [63:0] of the
1558 ///    destination. \n
1559 ///    Bit [1]=0: Bits [63:0] are copied from \a b to bits [127:64] of the
1560 ///    destination. \n
1561 ///    Bit [1]=1: Bits [127:64] are copied from \a b to bits [127:64] of the
1562 ///    destination. \n
1563 ///    Bit [2]=0: Bits [191:128] are copied from \a a to bits [191:128] of the
1564 ///    destination. \n
1565 ///    Bit [2]=1: Bits [255:192] are copied from \a a to bits [191:128] of the
1566 ///    destination. \n
1567 ///    Bit [3]=0: Bits [191:128] are copied from \a b to bits [255:192] of the
1568 ///    destination. \n
1569 ///    Bit [3]=1: Bits [255:192] are copied from \a b to bits [255:192] of the
1570 ///    destination.
1571 /// \returns A 256-bit vector of [4 x double] containing the shuffled values.
1572 #define _mm256_shuffle_pd(a, b, mask) \
1573   ((__m256d)__builtin_ia32_shufpd256((__v4df)(__m256d)(a), \
1574                                      (__v4df)(__m256d)(b), (int)(mask)))
1575 
1576 /* Compare */
1577 #define _CMP_EQ_OQ    0x00 /* Equal (ordered, non-signaling)  */
1578 #define _CMP_LT_OS    0x01 /* Less-than (ordered, signaling)  */
1579 #define _CMP_LE_OS    0x02 /* Less-than-or-equal (ordered, signaling)  */
1580 #define _CMP_UNORD_Q  0x03 /* Unordered (non-signaling)  */
1581 #define _CMP_NEQ_UQ   0x04 /* Not-equal (unordered, non-signaling)  */
1582 #define _CMP_NLT_US   0x05 /* Not-less-than (unordered, signaling)  */
1583 #define _CMP_NLE_US   0x06 /* Not-less-than-or-equal (unordered, signaling)  */
1584 #define _CMP_ORD_Q    0x07 /* Ordered (non-signaling)   */
1585 #define _CMP_EQ_UQ    0x08 /* Equal (unordered, non-signaling)  */
1586 #define _CMP_NGE_US   0x09 /* Not-greater-than-or-equal (unordered, signaling)  */
1587 #define _CMP_NGT_US   0x0a /* Not-greater-than (unordered, signaling)  */
1588 #define _CMP_FALSE_OQ 0x0b /* False (ordered, non-signaling)  */
1589 #define _CMP_NEQ_OQ   0x0c /* Not-equal (ordered, non-signaling)  */
1590 #define _CMP_GE_OS    0x0d /* Greater-than-or-equal (ordered, signaling)  */
1591 #define _CMP_GT_OS    0x0e /* Greater-than (ordered, signaling)  */
1592 #define _CMP_TRUE_UQ  0x0f /* True (unordered, non-signaling)  */
1593 #define _CMP_EQ_OS    0x10 /* Equal (ordered, signaling)  */
1594 #define _CMP_LT_OQ    0x11 /* Less-than (ordered, non-signaling)  */
1595 #define _CMP_LE_OQ    0x12 /* Less-than-or-equal (ordered, non-signaling)  */
1596 #define _CMP_UNORD_S  0x13 /* Unordered (signaling)  */
1597 #define _CMP_NEQ_US   0x14 /* Not-equal (unordered, signaling)  */
1598 #define _CMP_NLT_UQ   0x15 /* Not-less-than (unordered, non-signaling)  */
1599 #define _CMP_NLE_UQ   0x16 /* Not-less-than-or-equal (unordered, non-signaling)  */
1600 #define _CMP_ORD_S    0x17 /* Ordered (signaling)  */
1601 #define _CMP_EQ_US    0x18 /* Equal (unordered, signaling)  */
1602 #define _CMP_NGE_UQ   0x19 /* Not-greater-than-or-equal (unordered, non-signaling)  */
1603 #define _CMP_NGT_UQ   0x1a /* Not-greater-than (unordered, non-signaling)  */
1604 #define _CMP_FALSE_OS 0x1b /* False (ordered, signaling)  */
1605 #define _CMP_NEQ_OS   0x1c /* Not-equal (ordered, signaling)  */
1606 #define _CMP_GE_OQ    0x1d /* Greater-than-or-equal (ordered, non-signaling)  */
1607 #define _CMP_GT_OQ    0x1e /* Greater-than (ordered, non-signaling)  */
1608 #define _CMP_TRUE_US  0x1f /* True (unordered, signaling)  */
1609 
1610 /// Compares each of the corresponding double-precision values of two
1611 ///    128-bit vectors of [2 x double], using the operation specified by the
1612 ///    immediate integer operand.
1613 ///
1614 ///    Returns a [2 x double] vector consisting of two doubles corresponding to
1615 ///    the two comparison results: zero if the comparison is false, and all 1's
1616 ///    if the comparison is true.
1617 ///
1618 /// \headerfile <x86intrin.h>
1619 ///
1620 /// \code
1621 /// __m128d _mm_cmp_pd(__m128d a, __m128d b, const int c);
1622 /// \endcode
1623 ///
1624 /// This intrinsic corresponds to the <c> VCMPPD </c> instruction.
1625 ///
1626 /// \param a
1627 ///    A 128-bit vector of [2 x double].
1628 /// \param b
1629 ///    A 128-bit vector of [2 x double].
1630 /// \param c
1631 ///    An immediate integer operand, with bits [4:0] specifying which comparison
1632 ///    operation to use: \n
1633 ///    0x00: Equal (ordered, non-signaling) \n
1634 ///    0x01: Less-than (ordered, signaling) \n
1635 ///    0x02: Less-than-or-equal (ordered, signaling) \n
1636 ///    0x03: Unordered (non-signaling) \n
1637 ///    0x04: Not-equal (unordered, non-signaling) \n
1638 ///    0x05: Not-less-than (unordered, signaling) \n
1639 ///    0x06: Not-less-than-or-equal (unordered, signaling) \n
1640 ///    0x07: Ordered (non-signaling) \n
1641 ///    0x08: Equal (unordered, non-signaling) \n
1642 ///    0x09: Not-greater-than-or-equal (unordered, signaling) \n
1643 ///    0x0A: Not-greater-than (unordered, signaling) \n
1644 ///    0x0B: False (ordered, non-signaling) \n
1645 ///    0x0C: Not-equal (ordered, non-signaling) \n
1646 ///    0x0D: Greater-than-or-equal (ordered, signaling) \n
1647 ///    0x0E: Greater-than (ordered, signaling) \n
1648 ///    0x0F: True (unordered, non-signaling) \n
1649 ///    0x10: Equal (ordered, signaling) \n
1650 ///    0x11: Less-than (ordered, non-signaling) \n
1651 ///    0x12: Less-than-or-equal (ordered, non-signaling) \n
1652 ///    0x13: Unordered (signaling) \n
1653 ///    0x14: Not-equal (unordered, signaling) \n
1654 ///    0x15: Not-less-than (unordered, non-signaling) \n
1655 ///    0x16: Not-less-than-or-equal (unordered, non-signaling) \n
1656 ///    0x17: Ordered (signaling) \n
1657 ///    0x18: Equal (unordered, signaling) \n
1658 ///    0x19: Not-greater-than-or-equal (unordered, non-signaling) \n
1659 ///    0x1A: Not-greater-than (unordered, non-signaling) \n
1660 ///    0x1B: False (ordered, signaling) \n
1661 ///    0x1C: Not-equal (ordered, signaling) \n
1662 ///    0x1D: Greater-than-or-equal (ordered, non-signaling) \n
1663 ///    0x1E: Greater-than (ordered, non-signaling) \n
1664 ///    0x1F: True (unordered, signaling)
1665 /// \returns A 128-bit vector of [2 x double] containing the comparison results.
1666 #define _mm_cmp_pd(a, b, c) \
1667   ((__m128d)__builtin_ia32_cmppd((__v2df)(__m128d)(a), \
1668                                  (__v2df)(__m128d)(b), (c)))
1669 
1670 /// Compares each of the corresponding values of two 128-bit vectors of
1671 ///    [4 x float], using the operation specified by the immediate integer
1672 ///    operand.
1673 ///
1674 ///    Returns a [4 x float] vector consisting of four floats corresponding to
1675 ///    the four comparison results: zero if the comparison is false, and all 1's
1676 ///    if the comparison is true.
1677 ///
1678 /// \headerfile <x86intrin.h>
1679 ///
1680 /// \code
1681 /// __m128 _mm_cmp_ps(__m128 a, __m128 b, const int c);
1682 /// \endcode
1683 ///
1684 /// This intrinsic corresponds to the <c> VCMPPS </c> instruction.
1685 ///
1686 /// \param a
1687 ///    A 128-bit vector of [4 x float].
1688 /// \param b
1689 ///    A 128-bit vector of [4 x float].
1690 /// \param c
1691 ///    An immediate integer operand, with bits [4:0] specifying which comparison
1692 ///    operation to use: \n
1693 ///    0x00: Equal (ordered, non-signaling) \n
1694 ///    0x01: Less-than (ordered, signaling) \n
1695 ///    0x02: Less-than-or-equal (ordered, signaling) \n
1696 ///    0x03: Unordered (non-signaling) \n
1697 ///    0x04: Not-equal (unordered, non-signaling) \n
1698 ///    0x05: Not-less-than (unordered, signaling) \n
1699 ///    0x06: Not-less-than-or-equal (unordered, signaling) \n
1700 ///    0x07: Ordered (non-signaling) \n
1701 ///    0x08: Equal (unordered, non-signaling) \n
1702 ///    0x09: Not-greater-than-or-equal (unordered, signaling) \n
1703 ///    0x0A: Not-greater-than (unordered, signaling) \n
1704 ///    0x0B: False (ordered, non-signaling) \n
1705 ///    0x0C: Not-equal (ordered, non-signaling) \n
1706 ///    0x0D: Greater-than-or-equal (ordered, signaling) \n
1707 ///    0x0E: Greater-than (ordered, signaling) \n
1708 ///    0x0F: True (unordered, non-signaling) \n
1709 ///    0x10: Equal (ordered, signaling) \n
1710 ///    0x11: Less-than (ordered, non-signaling) \n
1711 ///    0x12: Less-than-or-equal (ordered, non-signaling) \n
1712 ///    0x13: Unordered (signaling) \n
1713 ///    0x14: Not-equal (unordered, signaling) \n
1714 ///    0x15: Not-less-than (unordered, non-signaling) \n
1715 ///    0x16: Not-less-than-or-equal (unordered, non-signaling) \n
1716 ///    0x17: Ordered (signaling) \n
1717 ///    0x18: Equal (unordered, signaling) \n
1718 ///    0x19: Not-greater-than-or-equal (unordered, non-signaling) \n
1719 ///    0x1A: Not-greater-than (unordered, non-signaling) \n
1720 ///    0x1B: False (ordered, signaling) \n
1721 ///    0x1C: Not-equal (ordered, signaling) \n
1722 ///    0x1D: Greater-than-or-equal (ordered, non-signaling) \n
1723 ///    0x1E: Greater-than (ordered, non-signaling) \n
1724 ///    0x1F: True (unordered, signaling)
1725 /// \returns A 128-bit vector of [4 x float] containing the comparison results.
1726 #define _mm_cmp_ps(a, b, c) \
1727   ((__m128)__builtin_ia32_cmpps((__v4sf)(__m128)(a), \
1728                                 (__v4sf)(__m128)(b), (c)))
1729 
1730 /// Compares each of the corresponding double-precision values of two
1731 ///    256-bit vectors of [4 x double], using the operation specified by the
1732 ///    immediate integer operand.
1733 ///
1734 ///    Returns a [4 x double] vector consisting of four doubles corresponding to
1735 ///    the four comparison results: zero if the comparison is false, and all 1's
1736 ///    if the comparison is true.
1737 ///
1738 /// \headerfile <x86intrin.h>
1739 ///
1740 /// \code
1741 /// __m256d _mm256_cmp_pd(__m256d a, __m256d b, const int c);
1742 /// \endcode
1743 ///
1744 /// This intrinsic corresponds to the <c> VCMPPD </c> instruction.
1745 ///
1746 /// \param a
1747 ///    A 256-bit vector of [4 x double].
1748 /// \param b
1749 ///    A 256-bit vector of [4 x double].
1750 /// \param c
1751 ///    An immediate integer operand, with bits [4:0] specifying which comparison
1752 ///    operation to use: \n
1753 ///    0x00: Equal (ordered, non-signaling) \n
1754 ///    0x01: Less-than (ordered, signaling) \n
1755 ///    0x02: Less-than-or-equal (ordered, signaling) \n
1756 ///    0x03: Unordered (non-signaling) \n
1757 ///    0x04: Not-equal (unordered, non-signaling) \n
1758 ///    0x05: Not-less-than (unordered, signaling) \n
1759 ///    0x06: Not-less-than-or-equal (unordered, signaling) \n
1760 ///    0x07: Ordered (non-signaling) \n
1761 ///    0x08: Equal (unordered, non-signaling) \n
1762 ///    0x09: Not-greater-than-or-equal (unordered, signaling) \n
1763 ///    0x0A: Not-greater-than (unordered, signaling) \n
1764 ///    0x0B: False (ordered, non-signaling) \n
1765 ///    0x0C: Not-equal (ordered, non-signaling) \n
1766 ///    0x0D: Greater-than-or-equal (ordered, signaling) \n
1767 ///    0x0E: Greater-than (ordered, signaling) \n
1768 ///    0x0F: True (unordered, non-signaling) \n
1769 ///    0x10: Equal (ordered, signaling) \n
1770 ///    0x11: Less-than (ordered, non-signaling) \n
1771 ///    0x12: Less-than-or-equal (ordered, non-signaling) \n
1772 ///    0x13: Unordered (signaling) \n
1773 ///    0x14: Not-equal (unordered, signaling) \n
1774 ///    0x15: Not-less-than (unordered, non-signaling) \n
1775 ///    0x16: Not-less-than-or-equal (unordered, non-signaling) \n
1776 ///    0x17: Ordered (signaling) \n
1777 ///    0x18: Equal (unordered, signaling) \n
1778 ///    0x19: Not-greater-than-or-equal (unordered, non-signaling) \n
1779 ///    0x1A: Not-greater-than (unordered, non-signaling) \n
1780 ///    0x1B: False (ordered, signaling) \n
1781 ///    0x1C: Not-equal (ordered, signaling) \n
1782 ///    0x1D: Greater-than-or-equal (ordered, non-signaling) \n
1783 ///    0x1E: Greater-than (ordered, non-signaling) \n
1784 ///    0x1F: True (unordered, signaling)
1785 /// \returns A 256-bit vector of [4 x double] containing the comparison results.
1786 #define _mm256_cmp_pd(a, b, c) \
1787   ((__m256d)__builtin_ia32_cmppd256((__v4df)(__m256d)(a), \
1788                                     (__v4df)(__m256d)(b), (c)))
1789 
1790 /// Compares each of the corresponding values of two 256-bit vectors of
1791 ///    [8 x float], using the operation specified by the immediate integer
1792 ///    operand.
1793 ///
1794 ///    Returns a [8 x float] vector consisting of eight floats corresponding to
1795 ///    the eight comparison results: zero if the comparison is false, and all
1796 ///    1's if the comparison is true.
1797 ///
1798 /// \headerfile <x86intrin.h>
1799 ///
1800 /// \code
1801 /// __m256 _mm256_cmp_ps(__m256 a, __m256 b, const int c);
1802 /// \endcode
1803 ///
1804 /// This intrinsic corresponds to the <c> VCMPPS </c> instruction.
1805 ///
1806 /// \param a
1807 ///    A 256-bit vector of [8 x float].
1808 /// \param b
1809 ///    A 256-bit vector of [8 x float].
1810 /// \param c
1811 ///    An immediate integer operand, with bits [4:0] specifying which comparison
1812 ///    operation to use: \n
1813 ///    0x00: Equal (ordered, non-signaling) \n
1814 ///    0x01: Less-than (ordered, signaling) \n
1815 ///    0x02: Less-than-or-equal (ordered, signaling) \n
1816 ///    0x03: Unordered (non-signaling) \n
1817 ///    0x04: Not-equal (unordered, non-signaling) \n
1818 ///    0x05: Not-less-than (unordered, signaling) \n
1819 ///    0x06: Not-less-than-or-equal (unordered, signaling) \n
1820 ///    0x07: Ordered (non-signaling) \n
1821 ///    0x08: Equal (unordered, non-signaling) \n
1822 ///    0x09: Not-greater-than-or-equal (unordered, signaling) \n
1823 ///    0x0A: Not-greater-than (unordered, signaling) \n
1824 ///    0x0B: False (ordered, non-signaling) \n
1825 ///    0x0C: Not-equal (ordered, non-signaling) \n
1826 ///    0x0D: Greater-than-or-equal (ordered, signaling) \n
1827 ///    0x0E: Greater-than (ordered, signaling) \n
1828 ///    0x0F: True (unordered, non-signaling) \n
1829 ///    0x10: Equal (ordered, signaling) \n
1830 ///    0x11: Less-than (ordered, non-signaling) \n
1831 ///    0x12: Less-than-or-equal (ordered, non-signaling) \n
1832 ///    0x13: Unordered (signaling) \n
1833 ///    0x14: Not-equal (unordered, signaling) \n
1834 ///    0x15: Not-less-than (unordered, non-signaling) \n
1835 ///    0x16: Not-less-than-or-equal (unordered, non-signaling) \n
1836 ///    0x17: Ordered (signaling) \n
1837 ///    0x18: Equal (unordered, signaling) \n
1838 ///    0x19: Not-greater-than-or-equal (unordered, non-signaling) \n
1839 ///    0x1A: Not-greater-than (unordered, non-signaling) \n
1840 ///    0x1B: False (ordered, signaling) \n
1841 ///    0x1C: Not-equal (ordered, signaling) \n
1842 ///    0x1D: Greater-than-or-equal (ordered, non-signaling) \n
1843 ///    0x1E: Greater-than (ordered, non-signaling) \n
1844 ///    0x1F: True (unordered, signaling)
1845 /// \returns A 256-bit vector of [8 x float] containing the comparison results.
1846 #define _mm256_cmp_ps(a, b, c) \
1847   ((__m256)__builtin_ia32_cmpps256((__v8sf)(__m256)(a), \
1848                                    (__v8sf)(__m256)(b), (c)))
1849 
1850 /// Compares each of the corresponding scalar double-precision values of
1851 ///    two 128-bit vectors of [2 x double], using the operation specified by the
1852 ///    immediate integer operand.
1853 ///
1854 ///    If the result is true, all 64 bits of the destination vector are set;
1855 ///    otherwise they are cleared.
1856 ///
1857 /// \headerfile <x86intrin.h>
1858 ///
1859 /// \code
1860 /// __m128d _mm_cmp_sd(__m128d a, __m128d b, const int c);
1861 /// \endcode
1862 ///
1863 /// This intrinsic corresponds to the <c> VCMPSD </c> instruction.
1864 ///
1865 /// \param a
1866 ///    A 128-bit vector of [2 x double].
1867 /// \param b
1868 ///    A 128-bit vector of [2 x double].
1869 /// \param c
1870 ///    An immediate integer operand, with bits [4:0] specifying which comparison
1871 ///    operation to use: \n
1872 ///    0x00: Equal (ordered, non-signaling) \n
1873 ///    0x01: Less-than (ordered, signaling) \n
1874 ///    0x02: Less-than-or-equal (ordered, signaling) \n
1875 ///    0x03: Unordered (non-signaling) \n
1876 ///    0x04: Not-equal (unordered, non-signaling) \n
1877 ///    0x05: Not-less-than (unordered, signaling) \n
1878 ///    0x06: Not-less-than-or-equal (unordered, signaling) \n
1879 ///    0x07: Ordered (non-signaling) \n
1880 ///    0x08: Equal (unordered, non-signaling) \n
1881 ///    0x09: Not-greater-than-or-equal (unordered, signaling) \n
1882 ///    0x0A: Not-greater-than (unordered, signaling) \n
1883 ///    0x0B: False (ordered, non-signaling) \n
1884 ///    0x0C: Not-equal (ordered, non-signaling) \n
1885 ///    0x0D: Greater-than-or-equal (ordered, signaling) \n
1886 ///    0x0E: Greater-than (ordered, signaling) \n
1887 ///    0x0F: True (unordered, non-signaling) \n
1888 ///    0x10: Equal (ordered, signaling) \n
1889 ///    0x11: Less-than (ordered, non-signaling) \n
1890 ///    0x12: Less-than-or-equal (ordered, non-signaling) \n
1891 ///    0x13: Unordered (signaling) \n
1892 ///    0x14: Not-equal (unordered, signaling) \n
1893 ///    0x15: Not-less-than (unordered, non-signaling) \n
1894 ///    0x16: Not-less-than-or-equal (unordered, non-signaling) \n
1895 ///    0x17: Ordered (signaling) \n
1896 ///    0x18: Equal (unordered, signaling) \n
1897 ///    0x19: Not-greater-than-or-equal (unordered, non-signaling) \n
1898 ///    0x1A: Not-greater-than (unordered, non-signaling) \n
1899 ///    0x1B: False (ordered, signaling) \n
1900 ///    0x1C: Not-equal (ordered, signaling) \n
1901 ///    0x1D: Greater-than-or-equal (ordered, non-signaling) \n
1902 ///    0x1E: Greater-than (ordered, non-signaling) \n
1903 ///    0x1F: True (unordered, signaling)
1904 /// \returns A 128-bit vector of [2 x double] containing the comparison results.
1905 #define _mm_cmp_sd(a, b, c) \
1906   ((__m128d)__builtin_ia32_cmpsd((__v2df)(__m128d)(a), \
1907                                  (__v2df)(__m128d)(b), (c)))
1908 
1909 /// Compares each of the corresponding scalar values of two 128-bit
1910 ///    vectors of [4 x float], using the operation specified by the immediate
1911 ///    integer operand.
1912 ///
1913 ///    If the result is true, all 32 bits of the destination vector are set;
1914 ///    otherwise they are cleared.
1915 ///
1916 /// \headerfile <x86intrin.h>
1917 ///
1918 /// \code
1919 /// __m128 _mm_cmp_ss(__m128 a, __m128 b, const int c);
1920 /// \endcode
1921 ///
1922 /// This intrinsic corresponds to the <c> VCMPSS </c> instruction.
1923 ///
1924 /// \param a
1925 ///    A 128-bit vector of [4 x float].
1926 /// \param b
1927 ///    A 128-bit vector of [4 x float].
1928 /// \param c
1929 ///    An immediate integer operand, with bits [4:0] specifying which comparison
1930 ///    operation to use: \n
1931 ///    0x00: Equal (ordered, non-signaling) \n
1932 ///    0x01: Less-than (ordered, signaling) \n
1933 ///    0x02: Less-than-or-equal (ordered, signaling) \n
1934 ///    0x03: Unordered (non-signaling) \n
1935 ///    0x04: Not-equal (unordered, non-signaling) \n
1936 ///    0x05: Not-less-than (unordered, signaling) \n
1937 ///    0x06: Not-less-than-or-equal (unordered, signaling) \n
1938 ///    0x07: Ordered (non-signaling) \n
1939 ///    0x08: Equal (unordered, non-signaling) \n
1940 ///    0x09: Not-greater-than-or-equal (unordered, signaling) \n
1941 ///    0x0A: Not-greater-than (unordered, signaling) \n
1942 ///    0x0B: False (ordered, non-signaling) \n
1943 ///    0x0C: Not-equal (ordered, non-signaling) \n
1944 ///    0x0D: Greater-than-or-equal (ordered, signaling) \n
1945 ///    0x0E: Greater-than (ordered, signaling) \n
1946 ///    0x0F: True (unordered, non-signaling) \n
1947 ///    0x10: Equal (ordered, signaling) \n
1948 ///    0x11: Less-than (ordered, non-signaling) \n
1949 ///    0x12: Less-than-or-equal (ordered, non-signaling) \n
1950 ///    0x13: Unordered (signaling) \n
1951 ///    0x14: Not-equal (unordered, signaling) \n
1952 ///    0x15: Not-less-than (unordered, non-signaling) \n
1953 ///    0x16: Not-less-than-or-equal (unordered, non-signaling) \n
1954 ///    0x17: Ordered (signaling) \n
1955 ///    0x18: Equal (unordered, signaling) \n
1956 ///    0x19: Not-greater-than-or-equal (unordered, non-signaling) \n
1957 ///    0x1A: Not-greater-than (unordered, non-signaling) \n
1958 ///    0x1B: False (ordered, signaling) \n
1959 ///    0x1C: Not-equal (ordered, signaling) \n
1960 ///    0x1D: Greater-than-or-equal (ordered, non-signaling) \n
1961 ///    0x1E: Greater-than (ordered, non-signaling) \n
1962 ///    0x1F: True (unordered, signaling)
1963 /// \returns A 128-bit vector of [4 x float] containing the comparison results.
1964 #define _mm_cmp_ss(a, b, c) \
1965   ((__m128)__builtin_ia32_cmpss((__v4sf)(__m128)(a), \
1966                                 (__v4sf)(__m128)(b), (c)))
1967 
1968 /// Takes a [8 x i32] vector and returns the vector element value
1969 ///    indexed by the immediate constant operand.
1970 ///
1971 /// \headerfile <x86intrin.h>
1972 ///
1973 /// \code
1974 /// int _mm256_extract_epi32(__m256i X, const int N);
1975 /// \endcode
1976 ///
1977 /// This intrinsic corresponds to the <c> VEXTRACTF128+COMPOSITE </c>
1978 ///   instruction.
1979 ///
1980 /// \param X
1981 ///    A 256-bit vector of [8 x i32].
1982 /// \param N
1983 ///    An immediate integer operand with bits [2:0] determining which vector
1984 ///    element is extracted and returned.
1985 /// \returns A 32-bit integer containing the extracted 32 bits of extended
1986 ///    packed data.
1987 #define _mm256_extract_epi32(X, N) \
1988   ((int)__builtin_ia32_vec_ext_v8si((__v8si)(__m256i)(X), (int)(N)))
1989 
1990 /// Takes a [16 x i16] vector and returns the vector element value
1991 ///    indexed by the immediate constant operand.
1992 ///
1993 /// \headerfile <x86intrin.h>
1994 ///
1995 /// \code
1996 /// int _mm256_extract_epi16(__m256i X, const int N);
1997 /// \endcode
1998 ///
1999 /// This intrinsic corresponds to the <c> VEXTRACTF128+COMPOSITE </c>
2000 ///   instruction.
2001 ///
2002 /// \param X
2003 ///    A 256-bit integer vector of [16 x i16].
2004 /// \param N
2005 ///    An immediate integer operand with bits [3:0] determining which vector
2006 ///    element is extracted and returned.
2007 /// \returns A 32-bit integer containing the extracted 16 bits of zero extended
2008 ///    packed data.
2009 #define _mm256_extract_epi16(X, N) \
2010   ((int)(unsigned short)__builtin_ia32_vec_ext_v16hi((__v16hi)(__m256i)(X), \
2011                                                      (int)(N)))
2012 
2013 /// Takes a [32 x i8] vector and returns the vector element value
2014 ///    indexed by the immediate constant operand.
2015 ///
2016 /// \headerfile <x86intrin.h>
2017 ///
2018 /// \code
2019 /// int _mm256_extract_epi8(__m256i X, const int N);
2020 /// \endcode
2021 ///
2022 /// This intrinsic corresponds to the <c> VEXTRACTF128+COMPOSITE </c>
2023 ///   instruction.
2024 ///
2025 /// \param X
2026 ///    A 256-bit integer vector of [32 x i8].
2027 /// \param N
2028 ///    An immediate integer operand with bits [4:0] determining which vector
2029 ///    element is extracted and returned.
2030 /// \returns A 32-bit integer containing the extracted 8 bits of zero extended
2031 ///    packed data.
2032 #define _mm256_extract_epi8(X, N) \
2033   ((int)(unsigned char)__builtin_ia32_vec_ext_v32qi((__v32qi)(__m256i)(X), \
2034                                                     (int)(N)))
2035 
2036 #ifdef __x86_64__
2037 /// Takes a [4 x i64] vector and returns the vector element value
2038 ///    indexed by the immediate constant operand.
2039 ///
2040 /// \headerfile <x86intrin.h>
2041 ///
2042 /// \code
2043 /// long long _mm256_extract_epi64(__m256i X, const int N);
2044 /// \endcode
2045 ///
2046 /// This intrinsic corresponds to the <c> VEXTRACTF128+COMPOSITE </c>
2047 ///   instruction.
2048 ///
2049 /// \param X
2050 ///    A 256-bit integer vector of [4 x i64].
2051 /// \param N
2052 ///    An immediate integer operand with bits [1:0] determining which vector
2053 ///    element is extracted and returned.
2054 /// \returns A 64-bit integer containing the extracted 64 bits of extended
2055 ///    packed data.
2056 #define _mm256_extract_epi64(X, N) \
2057   ((long long)__builtin_ia32_vec_ext_v4di((__v4di)(__m256i)(X), (int)(N)))
2058 #endif
2059 
2060 /// Takes a [8 x i32] vector and replaces the vector element value
2061 ///    indexed by the immediate constant operand by a new value. Returns the
2062 ///    modified vector.
2063 ///
2064 /// \headerfile <x86intrin.h>
2065 ///
2066 /// \code
2067 /// __m256i _mm256_insert_epi32(__m256i X, int I, const int N);
2068 /// \endcode
2069 ///
2070 /// This intrinsic corresponds to the <c> VINSERTF128+COMPOSITE </c>
2071 ///   instruction.
2072 ///
2073 /// \param X
2074 ///    A vector of [8 x i32] to be used by the insert operation.
2075 /// \param I
2076 ///    An integer value. The replacement value for the insert operation.
2077 /// \param N
2078 ///    An immediate integer specifying the index of the vector element to be
2079 ///    replaced.
2080 /// \returns A copy of vector \a X, after replacing its element indexed by
2081 ///    \a N with \a I.
2082 #define _mm256_insert_epi32(X, I, N) \
2083   ((__m256i)__builtin_ia32_vec_set_v8si((__v8si)(__m256i)(X), \
2084                                         (int)(I), (int)(N)))
2085 
2086 
2087 /// Takes a [16 x i16] vector and replaces the vector element value
2088 ///    indexed by the immediate constant operand with a new value. Returns the
2089 ///    modified vector.
2090 ///
2091 /// \headerfile <x86intrin.h>
2092 ///
2093 /// \code
2094 /// __m256i _mm256_insert_epi16(__m256i X, int I, const int N);
2095 /// \endcode
2096 ///
2097 /// This intrinsic corresponds to the <c> VINSERTF128+COMPOSITE </c>
2098 ///   instruction.
2099 ///
2100 /// \param X
2101 ///    A vector of [16 x i16] to be used by the insert operation.
2102 /// \param I
2103 ///    An i16 integer value. The replacement value for the insert operation.
2104 /// \param N
2105 ///    An immediate integer specifying the index of the vector element to be
2106 ///    replaced.
2107 /// \returns A copy of vector \a X, after replacing its element indexed by
2108 ///    \a N with \a I.
2109 #define _mm256_insert_epi16(X, I, N) \
2110   ((__m256i)__builtin_ia32_vec_set_v16hi((__v16hi)(__m256i)(X), \
2111                                          (int)(I), (int)(N)))
2112 
2113 /// Takes a [32 x i8] vector and replaces the vector element value
2114 ///    indexed by the immediate constant operand with a new value. Returns the
2115 ///    modified vector.
2116 ///
2117 /// \headerfile <x86intrin.h>
2118 ///
2119 /// \code
2120 /// __m256i _mm256_insert_epi8(__m256i X, int I, const int N);
2121 /// \endcode
2122 ///
2123 /// This intrinsic corresponds to the <c> VINSERTF128+COMPOSITE </c>
2124 ///   instruction.
2125 ///
2126 /// \param X
2127 ///    A vector of [32 x i8] to be used by the insert operation.
2128 /// \param I
2129 ///    An i8 integer value. The replacement value for the insert operation.
2130 /// \param N
2131 ///    An immediate integer specifying the index of the vector element to be
2132 ///    replaced.
2133 /// \returns A copy of vector \a X, after replacing its element indexed by
2134 ///    \a N with \a I.
2135 #define _mm256_insert_epi8(X, I, N) \
2136   ((__m256i)__builtin_ia32_vec_set_v32qi((__v32qi)(__m256i)(X), \
2137                                          (int)(I), (int)(N)))
2138 
2139 #ifdef __x86_64__
2140 /// Takes a [4 x i64] vector and replaces the vector element value
2141 ///    indexed by the immediate constant operand with a new value. Returns the
2142 ///    modified vector.
2143 ///
2144 /// \headerfile <x86intrin.h>
2145 ///
2146 /// \code
2147 /// __m256i _mm256_insert_epi64(__m256i X, int I, const int N);
2148 /// \endcode
2149 ///
2150 /// This intrinsic corresponds to the <c> VINSERTF128+COMPOSITE </c>
2151 ///   instruction.
2152 ///
2153 /// \param X
2154 ///    A vector of [4 x i64] to be used by the insert operation.
2155 /// \param I
2156 ///    A 64-bit integer value. The replacement value for the insert operation.
2157 /// \param N
2158 ///    An immediate integer specifying the index of the vector element to be
2159 ///    replaced.
2160 /// \returns A copy of vector \a X, after replacing its element indexed by
2161 ///     \a N with \a I.
2162 #define _mm256_insert_epi64(X, I, N) \
2163   ((__m256i)__builtin_ia32_vec_set_v4di((__v4di)(__m256i)(X), \
2164                                         (long long)(I), (int)(N)))
2165 #endif
2166 
2167 /* Conversion */
2168 /// Converts a vector of [4 x i32] into a vector of [4 x double].
2169 ///
2170 /// \headerfile <x86intrin.h>
2171 ///
2172 /// This intrinsic corresponds to the <c> VCVTDQ2PD </c> instruction.
2173 ///
2174 /// \param __a
2175 ///    A 128-bit integer vector of [4 x i32].
2176 /// \returns A 256-bit vector of [4 x double] containing the converted values.
2177 static __inline __m256d __DEFAULT_FN_ATTRS
2178 _mm256_cvtepi32_pd(__m128i __a)
2179 {
2180   return (__m256d)__builtin_convertvector((__v4si)__a, __v4df);
2181 }
2182 
2183 /// Converts a vector of [8 x i32] into a vector of [8 x float].
2184 ///
2185 /// \headerfile <x86intrin.h>
2186 ///
2187 /// This intrinsic corresponds to the <c> VCVTDQ2PS </c> instruction.
2188 ///
2189 /// \param __a
2190 ///    A 256-bit integer vector.
2191 /// \returns A 256-bit vector of [8 x float] containing the converted values.
2192 static __inline __m256 __DEFAULT_FN_ATTRS
2193 _mm256_cvtepi32_ps(__m256i __a)
2194 {
2195   return (__m256)__builtin_convertvector((__v8si)__a, __v8sf);
2196 }
2197 
2198 /// Converts a 256-bit vector of [4 x double] into a 128-bit vector of
2199 ///    [4 x float].
2200 ///
2201 /// \headerfile <x86intrin.h>
2202 ///
2203 /// This intrinsic corresponds to the <c> VCVTPD2PS </c> instruction.
2204 ///
2205 /// \param __a
2206 ///    A 256-bit vector of [4 x double].
2207 /// \returns A 128-bit vector of [4 x float] containing the converted values.
2208 static __inline __m128 __DEFAULT_FN_ATTRS
2209 _mm256_cvtpd_ps(__m256d __a)
2210 {
2211   return (__m128)__builtin_ia32_cvtpd2ps256((__v4df) __a);
2212 }
2213 
2214 /// Converts a vector of [8 x float] into a vector of [8 x i32].
2215 ///
2216 /// \headerfile <x86intrin.h>
2217 ///
2218 /// This intrinsic corresponds to the <c> VCVTPS2DQ </c> instruction.
2219 ///
2220 /// \param __a
2221 ///    A 256-bit vector of [8 x float].
2222 /// \returns A 256-bit integer vector containing the converted values.
2223 static __inline __m256i __DEFAULT_FN_ATTRS
2224 _mm256_cvtps_epi32(__m256 __a)
2225 {
2226   return (__m256i)__builtin_ia32_cvtps2dq256((__v8sf) __a);
2227 }
2228 
2229 /// Converts a 128-bit vector of [4 x float] into a 256-bit vector of [4
2230 ///    x double].
2231 ///
2232 /// \headerfile <x86intrin.h>
2233 ///
2234 /// This intrinsic corresponds to the <c> VCVTPS2PD </c> instruction.
2235 ///
2236 /// \param __a
2237 ///    A 128-bit vector of [4 x float].
2238 /// \returns A 256-bit vector of [4 x double] containing the converted values.
2239 static __inline __m256d __DEFAULT_FN_ATTRS
2240 _mm256_cvtps_pd(__m128 __a)
2241 {
2242   return (__m256d)__builtin_convertvector((__v4sf)__a, __v4df);
2243 }
2244 
2245 /// Converts a 256-bit vector of [4 x double] into a 128-bit vector of [4
2246 ///    x i32], truncating the result by rounding towards zero when it is
2247 ///    inexact.
2248 ///
2249 /// \headerfile <x86intrin.h>
2250 ///
2251 /// This intrinsic corresponds to the <c> VCVTTPD2DQ </c> instruction.
2252 ///
2253 /// \param __a
2254 ///    A 256-bit vector of [4 x double].
2255 /// \returns A 128-bit integer vector containing the converted values.
2256 static __inline __m128i __DEFAULT_FN_ATTRS
2257 _mm256_cvttpd_epi32(__m256d __a)
2258 {
2259   return (__m128i)__builtin_ia32_cvttpd2dq256((__v4df) __a);
2260 }
2261 
2262 /// Converts a 256-bit vector of [4 x double] into a 128-bit vector of [4
2263 ///    x i32]. When a conversion is inexact, the value returned is rounded
2264 ///    according to the rounding control bits in the MXCSR register.
2265 ///
2266 /// \headerfile <x86intrin.h>
2267 ///
2268 /// This intrinsic corresponds to the <c> VCVTPD2DQ </c> instruction.
2269 ///
2270 /// \param __a
2271 ///    A 256-bit vector of [4 x double].
2272 /// \returns A 128-bit integer vector containing the converted values.
2273 static __inline __m128i __DEFAULT_FN_ATTRS
2274 _mm256_cvtpd_epi32(__m256d __a)
2275 {
2276   return (__m128i)__builtin_ia32_cvtpd2dq256((__v4df) __a);
2277 }
2278 
2279 /// Converts a vector of [8 x float] into a vector of [8 x i32],
2280 ///    truncating the result by rounding towards zero when it is inexact.
2281 ///
2282 /// \headerfile <x86intrin.h>
2283 ///
2284 /// This intrinsic corresponds to the <c> VCVTTPS2DQ </c> instruction.
2285 ///
2286 /// \param __a
2287 ///    A 256-bit vector of [8 x float].
2288 /// \returns A 256-bit integer vector containing the converted values.
2289 static __inline __m256i __DEFAULT_FN_ATTRS
2290 _mm256_cvttps_epi32(__m256 __a)
2291 {
2292   return (__m256i)__builtin_ia32_cvttps2dq256((__v8sf) __a);
2293 }
2294 
2295 /// Returns the first element of the input vector of [4 x double].
2296 ///
2297 /// \headerfile <x86intrin.h>
2298 ///
2299 /// This intrinsic is a utility function and does not correspond to a specific
2300 ///    instruction.
2301 ///
2302 /// \param __a
2303 ///    A 256-bit vector of [4 x double].
2304 /// \returns A 64 bit double containing the first element of the input vector.
2305 static __inline double __DEFAULT_FN_ATTRS
2306 _mm256_cvtsd_f64(__m256d __a)
2307 {
2308  return __a[0];
2309 }
2310 
2311 /// Returns the first element of the input vector of [8 x i32].
2312 ///
2313 /// \headerfile <x86intrin.h>
2314 ///
2315 /// This intrinsic is a utility function and does not correspond to a specific
2316 ///    instruction.
2317 ///
2318 /// \param __a
2319 ///    A 256-bit vector of [8 x i32].
2320 /// \returns A 32 bit integer containing the first element of the input vector.
2321 static __inline int __DEFAULT_FN_ATTRS
2322 _mm256_cvtsi256_si32(__m256i __a)
2323 {
2324  __v8si __b = (__v8si)__a;
2325  return __b[0];
2326 }
2327 
2328 /// Returns the first element of the input vector of [8 x float].
2329 ///
2330 /// \headerfile <x86intrin.h>
2331 ///
2332 /// This intrinsic is a utility function and does not correspond to a specific
2333 ///    instruction.
2334 ///
2335 /// \param __a
2336 ///    A 256-bit vector of [8 x float].
2337 /// \returns A 32 bit float containing the first element of the input vector.
2338 static __inline float __DEFAULT_FN_ATTRS
2339 _mm256_cvtss_f32(__m256 __a)
2340 {
2341  return __a[0];
2342 }
2343 
2344 /* Vector replicate */
2345 /// Moves and duplicates odd-indexed values from a 256-bit vector of
2346 ///    [8 x float] to float values in a 256-bit vector of [8 x float].
2347 ///
2348 /// \headerfile <x86intrin.h>
2349 ///
2350 /// This intrinsic corresponds to the <c> VMOVSHDUP </c> instruction.
2351 ///
2352 /// \param __a
2353 ///    A 256-bit vector of [8 x float]. \n
2354 ///    Bits [255:224] of \a __a are written to bits [255:224] and [223:192] of
2355 ///    the return value. \n
2356 ///    Bits [191:160] of \a __a are written to bits [191:160] and [159:128] of
2357 ///    the return value. \n
2358 ///    Bits [127:96] of \a __a are written to bits [127:96] and [95:64] of the
2359 ///    return value. \n
2360 ///    Bits [63:32] of \a __a are written to bits [63:32] and [31:0] of the
2361 ///    return value.
2362 /// \returns A 256-bit vector of [8 x float] containing the moved and duplicated
2363 ///    values.
2364 static __inline __m256 __DEFAULT_FN_ATTRS
2365 _mm256_movehdup_ps(__m256 __a)
2366 {
2367   return __builtin_shufflevector((__v8sf)__a, (__v8sf)__a, 1, 1, 3, 3, 5, 5, 7, 7);
2368 }
2369 
2370 /// Moves and duplicates even-indexed values from a 256-bit vector of
2371 ///    [8 x float] to float values in a 256-bit vector of [8 x float].
2372 ///
2373 /// \headerfile <x86intrin.h>
2374 ///
2375 /// This intrinsic corresponds to the <c> VMOVSLDUP </c> instruction.
2376 ///
2377 /// \param __a
2378 ///    A 256-bit vector of [8 x float]. \n
2379 ///    Bits [223:192] of \a __a are written to bits [255:224] and [223:192] of
2380 ///    the return value. \n
2381 ///    Bits [159:128] of \a __a are written to bits [191:160] and [159:128] of
2382 ///    the return value. \n
2383 ///    Bits [95:64] of \a __a are written to bits [127:96] and [95:64] of the
2384 ///    return value. \n
2385 ///    Bits [31:0] of \a __a are written to bits [63:32] and [31:0] of the
2386 ///    return value.
2387 /// \returns A 256-bit vector of [8 x float] containing the moved and duplicated
2388 ///    values.
2389 static __inline __m256 __DEFAULT_FN_ATTRS
2390 _mm256_moveldup_ps(__m256 __a)
2391 {
2392   return __builtin_shufflevector((__v8sf)__a, (__v8sf)__a, 0, 0, 2, 2, 4, 4, 6, 6);
2393 }
2394 
2395 /// Moves and duplicates double-precision floating point values from a
2396 ///    256-bit vector of [4 x double] to double-precision values in a 256-bit
2397 ///    vector of [4 x double].
2398 ///
2399 /// \headerfile <x86intrin.h>
2400 ///
2401 /// This intrinsic corresponds to the <c> VMOVDDUP </c> instruction.
2402 ///
2403 /// \param __a
2404 ///    A 256-bit vector of [4 x double]. \n
2405 ///    Bits [63:0] of \a __a are written to bits [127:64] and [63:0] of the
2406 ///    return value. \n
2407 ///    Bits [191:128] of \a __a are written to bits [255:192] and [191:128] of
2408 ///    the return value.
2409 /// \returns A 256-bit vector of [4 x double] containing the moved and
2410 ///    duplicated values.
2411 static __inline __m256d __DEFAULT_FN_ATTRS
2412 _mm256_movedup_pd(__m256d __a)
2413 {
2414   return __builtin_shufflevector((__v4df)__a, (__v4df)__a, 0, 0, 2, 2);
2415 }
2416 
2417 /* Unpack and Interleave */
2418 /// Unpacks the odd-indexed vector elements from two 256-bit vectors of
2419 ///    [4 x double] and interleaves them into a 256-bit vector of [4 x double].
2420 ///
2421 /// \headerfile <x86intrin.h>
2422 ///
2423 /// This intrinsic corresponds to the <c> VUNPCKHPD </c> instruction.
2424 ///
2425 /// \param __a
2426 ///    A 256-bit floating-point vector of [4 x double]. \n
2427 ///    Bits [127:64] are written to bits [63:0] of the return value. \n
2428 ///    Bits [255:192] are written to bits [191:128] of the return value. \n
2429 /// \param __b
2430 ///    A 256-bit floating-point vector of [4 x double]. \n
2431 ///    Bits [127:64] are written to bits [127:64] of the return value. \n
2432 ///    Bits [255:192] are written to bits [255:192] of the return value. \n
2433 /// \returns A 256-bit vector of [4 x double] containing the interleaved values.
2434 static __inline __m256d __DEFAULT_FN_ATTRS
2435 _mm256_unpackhi_pd(__m256d __a, __m256d __b)
2436 {
2437   return __builtin_shufflevector((__v4df)__a, (__v4df)__b, 1, 5, 1+2, 5+2);
2438 }
2439 
2440 /// Unpacks the even-indexed vector elements from two 256-bit vectors of
2441 ///    [4 x double] and interleaves them into a 256-bit vector of [4 x double].
2442 ///
2443 /// \headerfile <x86intrin.h>
2444 ///
2445 /// This intrinsic corresponds to the <c> VUNPCKLPD </c> instruction.
2446 ///
2447 /// \param __a
2448 ///    A 256-bit floating-point vector of [4 x double]. \n
2449 ///    Bits [63:0] are written to bits [63:0] of the return value. \n
2450 ///    Bits [191:128] are written to bits [191:128] of the return value.
2451 /// \param __b
2452 ///    A 256-bit floating-point vector of [4 x double]. \n
2453 ///    Bits [63:0] are written to bits [127:64] of the return value. \n
2454 ///    Bits [191:128] are written to bits [255:192] of the return value. \n
2455 /// \returns A 256-bit vector of [4 x double] containing the interleaved values.
2456 static __inline __m256d __DEFAULT_FN_ATTRS
2457 _mm256_unpacklo_pd(__m256d __a, __m256d __b)
2458 {
2459   return __builtin_shufflevector((__v4df)__a, (__v4df)__b, 0, 4, 0+2, 4+2);
2460 }
2461 
2462 /// Unpacks the 32-bit vector elements 2, 3, 6 and 7 from each of the
2463 ///    two 256-bit vectors of [8 x float] and interleaves them into a 256-bit
2464 ///    vector of [8 x float].
2465 ///
2466 /// \headerfile <x86intrin.h>
2467 ///
2468 /// This intrinsic corresponds to the <c> VUNPCKHPS </c> instruction.
2469 ///
2470 /// \param __a
2471 ///    A 256-bit vector of [8 x float]. \n
2472 ///    Bits [95:64] are written to bits [31:0] of the return value. \n
2473 ///    Bits [127:96] are written to bits [95:64] of the return value. \n
2474 ///    Bits [223:192] are written to bits [159:128] of the return value. \n
2475 ///    Bits [255:224] are written to bits [223:192] of the return value.
2476 /// \param __b
2477 ///    A 256-bit vector of [8 x float]. \n
2478 ///    Bits [95:64] are written to bits [63:32] of the return value. \n
2479 ///    Bits [127:96] are written to bits [127:96] of the return value. \n
2480 ///    Bits [223:192] are written to bits [191:160] of the return value. \n
2481 ///    Bits [255:224] are written to bits [255:224] of the return value.
2482 /// \returns A 256-bit vector of [8 x float] containing the interleaved values.
2483 static __inline __m256 __DEFAULT_FN_ATTRS
2484 _mm256_unpackhi_ps(__m256 __a, __m256 __b)
2485 {
2486   return __builtin_shufflevector((__v8sf)__a, (__v8sf)__b, 2, 10, 2+1, 10+1, 6, 14, 6+1, 14+1);
2487 }
2488 
2489 /// Unpacks the 32-bit vector elements 0, 1, 4 and 5 from each of the
2490 ///    two 256-bit vectors of [8 x float] and interleaves them into a 256-bit
2491 ///    vector of [8 x float].
2492 ///
2493 /// \headerfile <x86intrin.h>
2494 ///
2495 /// This intrinsic corresponds to the <c> VUNPCKLPS </c> instruction.
2496 ///
2497 /// \param __a
2498 ///    A 256-bit vector of [8 x float]. \n
2499 ///    Bits [31:0] are written to bits [31:0] of the return value. \n
2500 ///    Bits [63:32] are written to bits [95:64] of the return value. \n
2501 ///    Bits [159:128] are written to bits [159:128] of the return value. \n
2502 ///    Bits [191:160] are written to bits [223:192] of the return value.
2503 /// \param __b
2504 ///    A 256-bit vector of [8 x float]. \n
2505 ///    Bits [31:0] are written to bits [63:32] of the return value. \n
2506 ///    Bits [63:32] are written to bits [127:96] of the return value. \n
2507 ///    Bits [159:128] are written to bits [191:160] of the return value. \n
2508 ///    Bits [191:160] are written to bits [255:224] of the return value.
2509 /// \returns A 256-bit vector of [8 x float] containing the interleaved values.
2510 static __inline __m256 __DEFAULT_FN_ATTRS
2511 _mm256_unpacklo_ps(__m256 __a, __m256 __b)
2512 {
2513   return __builtin_shufflevector((__v8sf)__a, (__v8sf)__b, 0, 8, 0+1, 8+1, 4, 12, 4+1, 12+1);
2514 }
2515 
2516 /* Bit Test */
2517 /// Given two 128-bit floating-point vectors of [2 x double], perform an
2518 ///    element-by-element comparison of the double-precision element in the
2519 ///    first source vector and the corresponding element in the second source
2520 ///    vector.
2521 ///
2522 ///    The EFLAGS register is updated as follows: \n
2523 ///    If there is at least one pair of double-precision elements where the
2524 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2525 ///    ZF flag is set to 1. \n
2526 ///    If there is at least one pair of double-precision elements where the
2527 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2528 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2529 ///    This intrinsic returns the value of the ZF flag.
2530 ///
2531 /// \headerfile <x86intrin.h>
2532 ///
2533 /// This intrinsic corresponds to the <c> VTESTPD </c> instruction.
2534 ///
2535 /// \param __a
2536 ///    A 128-bit vector of [2 x double].
2537 /// \param __b
2538 ///    A 128-bit vector of [2 x double].
2539 /// \returns the ZF flag in the EFLAGS register.
2540 static __inline int __DEFAULT_FN_ATTRS128
2541 _mm_testz_pd(__m128d __a, __m128d __b)
2542 {
2543   return __builtin_ia32_vtestzpd((__v2df)__a, (__v2df)__b);
2544 }
2545 
2546 /// Given two 128-bit floating-point vectors of [2 x double], perform an
2547 ///    element-by-element comparison of the double-precision element in the
2548 ///    first source vector and the corresponding element in the second source
2549 ///    vector.
2550 ///
2551 ///    The EFLAGS register is updated as follows: \n
2552 ///    If there is at least one pair of double-precision elements where the
2553 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2554 ///    ZF flag is set to 1. \n
2555 ///    If there is at least one pair of double-precision elements where the
2556 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2557 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2558 ///    This intrinsic returns the value of the CF flag.
2559 ///
2560 /// \headerfile <x86intrin.h>
2561 ///
2562 /// This intrinsic corresponds to the <c> VTESTPD </c> instruction.
2563 ///
2564 /// \param __a
2565 ///    A 128-bit vector of [2 x double].
2566 /// \param __b
2567 ///    A 128-bit vector of [2 x double].
2568 /// \returns the CF flag in the EFLAGS register.
2569 static __inline int __DEFAULT_FN_ATTRS128
2570 _mm_testc_pd(__m128d __a, __m128d __b)
2571 {
2572   return __builtin_ia32_vtestcpd((__v2df)__a, (__v2df)__b);
2573 }
2574 
2575 /// Given two 128-bit floating-point vectors of [2 x double], perform an
2576 ///    element-by-element comparison of the double-precision element in the
2577 ///    first source vector and the corresponding element in the second source
2578 ///    vector.
2579 ///
2580 ///    The EFLAGS register is updated as follows: \n
2581 ///    If there is at least one pair of double-precision elements where the
2582 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2583 ///    ZF flag is set to 1. \n
2584 ///    If there is at least one pair of double-precision elements where the
2585 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2586 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2587 ///    This intrinsic returns 1 if both the ZF and CF flags are set to 0,
2588 ///    otherwise it returns 0.
2589 ///
2590 /// \headerfile <x86intrin.h>
2591 ///
2592 /// This intrinsic corresponds to the <c> VTESTPD </c> instruction.
2593 ///
2594 /// \param __a
2595 ///    A 128-bit vector of [2 x double].
2596 /// \param __b
2597 ///    A 128-bit vector of [2 x double].
2598 /// \returns 1 if both the ZF and CF flags are set to 0, otherwise returns 0.
2599 static __inline int __DEFAULT_FN_ATTRS128
2600 _mm_testnzc_pd(__m128d __a, __m128d __b)
2601 {
2602   return __builtin_ia32_vtestnzcpd((__v2df)__a, (__v2df)__b);
2603 }
2604 
2605 /// Given two 128-bit floating-point vectors of [4 x float], perform an
2606 ///    element-by-element comparison of the single-precision element in the
2607 ///    first source vector and the corresponding element in the second source
2608 ///    vector.
2609 ///
2610 ///    The EFLAGS register is updated as follows: \n
2611 ///    If there is at least one pair of single-precision elements where the
2612 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2613 ///    ZF flag is set to 1. \n
2614 ///    If there is at least one pair of single-precision elements where the
2615 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2616 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2617 ///    This intrinsic returns the value of the ZF flag.
2618 ///
2619 /// \headerfile <x86intrin.h>
2620 ///
2621 /// This intrinsic corresponds to the <c> VTESTPS </c> instruction.
2622 ///
2623 /// \param __a
2624 ///    A 128-bit vector of [4 x float].
2625 /// \param __b
2626 ///    A 128-bit vector of [4 x float].
2627 /// \returns the ZF flag.
2628 static __inline int __DEFAULT_FN_ATTRS128
2629 _mm_testz_ps(__m128 __a, __m128 __b)
2630 {
2631   return __builtin_ia32_vtestzps((__v4sf)__a, (__v4sf)__b);
2632 }
2633 
2634 /// Given two 128-bit floating-point vectors of [4 x float], perform an
2635 ///    element-by-element comparison of the single-precision element in the
2636 ///    first source vector and the corresponding element in the second source
2637 ///    vector.
2638 ///
2639 ///    The EFLAGS register is updated as follows: \n
2640 ///    If there is at least one pair of single-precision elements where the
2641 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2642 ///    ZF flag is set to 1. \n
2643 ///    If there is at least one pair of single-precision elements where the
2644 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2645 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2646 ///    This intrinsic returns the value of the CF flag.
2647 ///
2648 /// \headerfile <x86intrin.h>
2649 ///
2650 /// This intrinsic corresponds to the <c> VTESTPS </c> instruction.
2651 ///
2652 /// \param __a
2653 ///    A 128-bit vector of [4 x float].
2654 /// \param __b
2655 ///    A 128-bit vector of [4 x float].
2656 /// \returns the CF flag.
2657 static __inline int __DEFAULT_FN_ATTRS128
2658 _mm_testc_ps(__m128 __a, __m128 __b)
2659 {
2660   return __builtin_ia32_vtestcps((__v4sf)__a, (__v4sf)__b);
2661 }
2662 
2663 /// Given two 128-bit floating-point vectors of [4 x float], perform an
2664 ///    element-by-element comparison of the single-precision element in the
2665 ///    first source vector and the corresponding element in the second source
2666 ///    vector.
2667 ///
2668 ///    The EFLAGS register is updated as follows: \n
2669 ///    If there is at least one pair of single-precision elements where the
2670 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2671 ///    ZF flag is set to 1. \n
2672 ///    If there is at least one pair of single-precision elements where the
2673 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2674 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2675 ///    This intrinsic returns 1 if both the ZF and CF flags are set to 0,
2676 ///    otherwise it returns 0.
2677 ///
2678 /// \headerfile <x86intrin.h>
2679 ///
2680 /// This intrinsic corresponds to the <c> VTESTPS </c> instruction.
2681 ///
2682 /// \param __a
2683 ///    A 128-bit vector of [4 x float].
2684 /// \param __b
2685 ///    A 128-bit vector of [4 x float].
2686 /// \returns 1 if both the ZF and CF flags are set to 0, otherwise returns 0.
2687 static __inline int __DEFAULT_FN_ATTRS128
2688 _mm_testnzc_ps(__m128 __a, __m128 __b)
2689 {
2690   return __builtin_ia32_vtestnzcps((__v4sf)__a, (__v4sf)__b);
2691 }
2692 
2693 /// Given two 256-bit floating-point vectors of [4 x double], perform an
2694 ///    element-by-element comparison of the double-precision elements in the
2695 ///    first source vector and the corresponding elements in the second source
2696 ///    vector.
2697 ///
2698 ///    The EFLAGS register is updated as follows: \n
2699 ///    If there is at least one pair of double-precision elements where the
2700 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2701 ///    ZF flag is set to 1. \n
2702 ///    If there is at least one pair of double-precision elements where the
2703 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2704 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2705 ///    This intrinsic returns the value of the ZF flag.
2706 ///
2707 /// \headerfile <x86intrin.h>
2708 ///
2709 /// This intrinsic corresponds to the <c> VTESTPD </c> instruction.
2710 ///
2711 /// \param __a
2712 ///    A 256-bit vector of [4 x double].
2713 /// \param __b
2714 ///    A 256-bit vector of [4 x double].
2715 /// \returns the ZF flag.
2716 static __inline int __DEFAULT_FN_ATTRS
2717 _mm256_testz_pd(__m256d __a, __m256d __b)
2718 {
2719   return __builtin_ia32_vtestzpd256((__v4df)__a, (__v4df)__b);
2720 }
2721 
2722 /// Given two 256-bit floating-point vectors of [4 x double], perform an
2723 ///    element-by-element comparison of the double-precision elements in the
2724 ///    first source vector and the corresponding elements in the second source
2725 ///    vector.
2726 ///
2727 ///    The EFLAGS register is updated as follows: \n
2728 ///    If there is at least one pair of double-precision elements where the
2729 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2730 ///    ZF flag is set to 1. \n
2731 ///    If there is at least one pair of double-precision elements where the
2732 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2733 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2734 ///    This intrinsic returns the value of the CF flag.
2735 ///
2736 /// \headerfile <x86intrin.h>
2737 ///
2738 /// This intrinsic corresponds to the <c> VTESTPD </c> instruction.
2739 ///
2740 /// \param __a
2741 ///    A 256-bit vector of [4 x double].
2742 /// \param __b
2743 ///    A 256-bit vector of [4 x double].
2744 /// \returns the CF flag.
2745 static __inline int __DEFAULT_FN_ATTRS
2746 _mm256_testc_pd(__m256d __a, __m256d __b)
2747 {
2748   return __builtin_ia32_vtestcpd256((__v4df)__a, (__v4df)__b);
2749 }
2750 
2751 /// Given two 256-bit floating-point vectors of [4 x double], perform an
2752 ///    element-by-element comparison of the double-precision elements in the
2753 ///    first source vector and the corresponding elements in the second source
2754 ///    vector.
2755 ///
2756 ///    The EFLAGS register is updated as follows: \n
2757 ///    If there is at least one pair of double-precision elements where the
2758 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2759 ///    ZF flag is set to 1. \n
2760 ///    If there is at least one pair of double-precision elements where the
2761 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2762 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2763 ///    This intrinsic returns 1 if both the ZF and CF flags are set to 0,
2764 ///    otherwise it returns 0.
2765 ///
2766 /// \headerfile <x86intrin.h>
2767 ///
2768 /// This intrinsic corresponds to the <c> VTESTPD </c> instruction.
2769 ///
2770 /// \param __a
2771 ///    A 256-bit vector of [4 x double].
2772 /// \param __b
2773 ///    A 256-bit vector of [4 x double].
2774 /// \returns 1 if both the ZF and CF flags are set to 0, otherwise returns 0.
2775 static __inline int __DEFAULT_FN_ATTRS
2776 _mm256_testnzc_pd(__m256d __a, __m256d __b)
2777 {
2778   return __builtin_ia32_vtestnzcpd256((__v4df)__a, (__v4df)__b);
2779 }
2780 
2781 /// Given two 256-bit floating-point vectors of [8 x float], perform an
2782 ///    element-by-element comparison of the single-precision element in the
2783 ///    first source vector and the corresponding element in the second source
2784 ///    vector.
2785 ///
2786 ///    The EFLAGS register is updated as follows: \n
2787 ///    If there is at least one pair of single-precision elements where the
2788 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2789 ///    ZF flag is set to 1. \n
2790 ///    If there is at least one pair of single-precision elements where the
2791 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2792 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2793 ///    This intrinsic returns the value of the ZF flag.
2794 ///
2795 /// \headerfile <x86intrin.h>
2796 ///
2797 /// This intrinsic corresponds to the <c> VTESTPS </c> instruction.
2798 ///
2799 /// \param __a
2800 ///    A 256-bit vector of [8 x float].
2801 /// \param __b
2802 ///    A 256-bit vector of [8 x float].
2803 /// \returns the ZF flag.
2804 static __inline int __DEFAULT_FN_ATTRS
2805 _mm256_testz_ps(__m256 __a, __m256 __b)
2806 {
2807   return __builtin_ia32_vtestzps256((__v8sf)__a, (__v8sf)__b);
2808 }
2809 
2810 /// Given two 256-bit floating-point vectors of [8 x float], perform an
2811 ///    element-by-element comparison of the single-precision element in the
2812 ///    first source vector and the corresponding element in the second source
2813 ///    vector.
2814 ///
2815 ///    The EFLAGS register is updated as follows: \n
2816 ///    If there is at least one pair of single-precision elements where the
2817 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2818 ///    ZF flag is set to 1. \n
2819 ///    If there is at least one pair of single-precision elements where the
2820 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2821 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2822 ///    This intrinsic returns the value of the CF flag.
2823 ///
2824 /// \headerfile <x86intrin.h>
2825 ///
2826 /// This intrinsic corresponds to the <c> VTESTPS </c> instruction.
2827 ///
2828 /// \param __a
2829 ///    A 256-bit vector of [8 x float].
2830 /// \param __b
2831 ///    A 256-bit vector of [8 x float].
2832 /// \returns the CF flag.
2833 static __inline int __DEFAULT_FN_ATTRS
2834 _mm256_testc_ps(__m256 __a, __m256 __b)
2835 {
2836   return __builtin_ia32_vtestcps256((__v8sf)__a, (__v8sf)__b);
2837 }
2838 
2839 /// Given two 256-bit floating-point vectors of [8 x float], perform an
2840 ///    element-by-element comparison of the single-precision elements in the
2841 ///    first source vector and the corresponding elements in the second source
2842 ///    vector.
2843 ///
2844 ///    The EFLAGS register is updated as follows: \n
2845 ///    If there is at least one pair of single-precision elements where the
2846 ///    sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the
2847 ///    ZF flag is set to 1. \n
2848 ///    If there is at least one pair of single-precision elements where the
2849 ///    sign-bit of the first element is 0 and the sign-bit of the second element
2850 ///    is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n
2851 ///    This intrinsic returns 1 if both the ZF and CF flags are set to 0,
2852 ///    otherwise it returns 0.
2853 ///
2854 /// \headerfile <x86intrin.h>
2855 ///
2856 /// This intrinsic corresponds to the <c> VTESTPS </c> instruction.
2857 ///
2858 /// \param __a
2859 ///    A 256-bit vector of [8 x float].
2860 /// \param __b
2861 ///    A 256-bit vector of [8 x float].
2862 /// \returns 1 if both the ZF and CF flags are set to 0, otherwise returns 0.
2863 static __inline int __DEFAULT_FN_ATTRS
2864 _mm256_testnzc_ps(__m256 __a, __m256 __b)
2865 {
2866   return __builtin_ia32_vtestnzcps256((__v8sf)__a, (__v8sf)__b);
2867 }
2868 
2869 /// Given two 256-bit integer vectors, perform a bit-by-bit comparison
2870 ///    of the two source vectors.
2871 ///
2872 ///    The EFLAGS register is updated as follows: \n
2873 ///    If there is at least one pair of bits where both bits are 1, the ZF flag
2874 ///    is set to 0. Otherwise the ZF flag is set to 1. \n
2875 ///    If there is at least one pair of bits where the bit from the first source
2876 ///    vector is 0 and the bit from the second source vector is 1, the CF flag
2877 ///    is set to 0. Otherwise the CF flag is set to 1. \n
2878 ///    This intrinsic returns the value of the ZF flag.
2879 ///
2880 /// \headerfile <x86intrin.h>
2881 ///
2882 /// This intrinsic corresponds to the <c> VPTEST </c> instruction.
2883 ///
2884 /// \param __a
2885 ///    A 256-bit integer vector.
2886 /// \param __b
2887 ///    A 256-bit integer vector.
2888 /// \returns the ZF flag.
2889 static __inline int __DEFAULT_FN_ATTRS
2890 _mm256_testz_si256(__m256i __a, __m256i __b)
2891 {
2892   return __builtin_ia32_ptestz256((__v4di)__a, (__v4di)__b);
2893 }
2894 
2895 /// Given two 256-bit integer vectors, perform a bit-by-bit comparison
2896 ///    of the two source vectors.
2897 ///
2898 ///    The EFLAGS register is updated as follows: \n
2899 ///    If there is at least one pair of bits where both bits are 1, the ZF flag
2900 ///    is set to 0. Otherwise the ZF flag is set to 1. \n
2901 ///    If there is at least one pair of bits where the bit from the first source
2902 ///    vector is 0 and the bit from the second source vector is 1, the CF flag
2903 ///    is set to 0. Otherwise the CF flag is set to 1. \n
2904 ///    This intrinsic returns the value of the CF flag.
2905 ///
2906 /// \headerfile <x86intrin.h>
2907 ///
2908 /// This intrinsic corresponds to the <c> VPTEST </c> instruction.
2909 ///
2910 /// \param __a
2911 ///    A 256-bit integer vector.
2912 /// \param __b
2913 ///    A 256-bit integer vector.
2914 /// \returns the CF flag.
2915 static __inline int __DEFAULT_FN_ATTRS
2916 _mm256_testc_si256(__m256i __a, __m256i __b)
2917 {
2918   return __builtin_ia32_ptestc256((__v4di)__a, (__v4di)__b);
2919 }
2920 
2921 /// Given two 256-bit integer vectors, perform a bit-by-bit comparison
2922 ///    of the two source vectors.
2923 ///
2924 ///    The EFLAGS register is updated as follows: \n
2925 ///    If there is at least one pair of bits where both bits are 1, the ZF flag
2926 ///    is set to 0. Otherwise the ZF flag is set to 1. \n
2927 ///    If there is at least one pair of bits where the bit from the first source
2928 ///    vector is 0 and the bit from the second source vector is 1, the CF flag
2929 ///    is set to 0. Otherwise the CF flag is set to 1. \n
2930 ///    This intrinsic returns 1 if both the ZF and CF flags are set to 0,
2931 ///    otherwise it returns 0.
2932 ///
2933 /// \headerfile <x86intrin.h>
2934 ///
2935 /// This intrinsic corresponds to the <c> VPTEST </c> instruction.
2936 ///
2937 /// \param __a
2938 ///    A 256-bit integer vector.
2939 /// \param __b
2940 ///    A 256-bit integer vector.
2941 /// \returns 1 if both the ZF and CF flags are set to 0, otherwise returns 0.
2942 static __inline int __DEFAULT_FN_ATTRS
2943 _mm256_testnzc_si256(__m256i __a, __m256i __b)
2944 {
2945   return __builtin_ia32_ptestnzc256((__v4di)__a, (__v4di)__b);
2946 }
2947 
2948 /* Vector extract sign mask */
2949 /// Extracts the sign bits of double-precision floating point elements
2950 ///    in a 256-bit vector of [4 x double] and writes them to the lower order
2951 ///    bits of the return value.
2952 ///
2953 /// \headerfile <x86intrin.h>
2954 ///
2955 /// This intrinsic corresponds to the <c> VMOVMSKPD </c> instruction.
2956 ///
2957 /// \param __a
2958 ///    A 256-bit vector of [4 x double] containing the double-precision
2959 ///    floating point values with sign bits to be extracted.
2960 /// \returns The sign bits from the operand, written to bits [3:0].
2961 static __inline int __DEFAULT_FN_ATTRS
2962 _mm256_movemask_pd(__m256d __a)
2963 {
2964   return __builtin_ia32_movmskpd256((__v4df)__a);
2965 }
2966 
2967 /// Extracts the sign bits of single-precision floating point elements
2968 ///    in a 256-bit vector of [8 x float] and writes them to the lower order
2969 ///    bits of the return value.
2970 ///
2971 /// \headerfile <x86intrin.h>
2972 ///
2973 /// This intrinsic corresponds to the <c> VMOVMSKPS </c> instruction.
2974 ///
2975 /// \param __a
2976 ///    A 256-bit vector of [8 x float] containing the single-precision floating
2977 ///    point values with sign bits to be extracted.
2978 /// \returns The sign bits from the operand, written to bits [7:0].
2979 static __inline int __DEFAULT_FN_ATTRS
2980 _mm256_movemask_ps(__m256 __a)
2981 {
2982   return __builtin_ia32_movmskps256((__v8sf)__a);
2983 }
2984 
2985 /* Vector __zero */
2986 /// Zeroes the contents of all XMM or YMM registers.
2987 ///
2988 /// \headerfile <x86intrin.h>
2989 ///
2990 /// This intrinsic corresponds to the <c> VZEROALL </c> instruction.
2991 static __inline void __attribute__((__always_inline__, __nodebug__, __target__("avx")))
2992 _mm256_zeroall(void)
2993 {
2994   __builtin_ia32_vzeroall();
2995 }
2996 
2997 /// Zeroes the upper 128 bits (bits 255:128) of all YMM registers.
2998 ///
2999 /// \headerfile <x86intrin.h>
3000 ///
3001 /// This intrinsic corresponds to the <c> VZEROUPPER </c> instruction.
3002 static __inline void __attribute__((__always_inline__, __nodebug__, __target__("avx")))
3003 _mm256_zeroupper(void)
3004 {
3005   __builtin_ia32_vzeroupper();
3006 }
3007 
3008 /* Vector load with broadcast */
3009 /// Loads a scalar single-precision floating point value from the
3010 ///    specified address pointed to by \a __a and broadcasts it to the elements
3011 ///    of a [4 x float] vector.
3012 ///
3013 /// \headerfile <x86intrin.h>
3014 ///
3015 /// This intrinsic corresponds to the <c> VBROADCASTSS </c> instruction.
3016 ///
3017 /// \param __a
3018 ///    The single-precision floating point value to be broadcast.
3019 /// \returns A 128-bit vector of [4 x float] whose 32-bit elements are set
3020 ///    equal to the broadcast value.
3021 static __inline __m128 __DEFAULT_FN_ATTRS128
3022 _mm_broadcast_ss(float const *__a)
3023 {
3024   struct __mm_broadcast_ss_struct {
3025     float __f;
3026   } __attribute__((__packed__, __may_alias__));
3027   float __f = ((const struct __mm_broadcast_ss_struct*)__a)->__f;
3028   return __extension__ (__m128){ __f, __f, __f, __f };
3029 }
3030 
3031 /// Loads a scalar double-precision floating point value from the
3032 ///    specified address pointed to by \a __a and broadcasts it to the elements
3033 ///    of a [4 x double] vector.
3034 ///
3035 /// \headerfile <x86intrin.h>
3036 ///
3037 /// This intrinsic corresponds to the <c> VBROADCASTSD </c> instruction.
3038 ///
3039 /// \param __a
3040 ///    The double-precision floating point value to be broadcast.
3041 /// \returns A 256-bit vector of [4 x double] whose 64-bit elements are set
3042 ///    equal to the broadcast value.
3043 static __inline __m256d __DEFAULT_FN_ATTRS
3044 _mm256_broadcast_sd(double const *__a)
3045 {
3046   struct __mm256_broadcast_sd_struct {
3047     double __d;
3048   } __attribute__((__packed__, __may_alias__));
3049   double __d = ((const struct __mm256_broadcast_sd_struct*)__a)->__d;
3050   return __extension__ (__m256d)(__v4df){ __d, __d, __d, __d };
3051 }
3052 
3053 /// Loads a scalar single-precision floating point value from the
3054 ///    specified address pointed to by \a __a and broadcasts it to the elements
3055 ///    of a [8 x float] vector.
3056 ///
3057 /// \headerfile <x86intrin.h>
3058 ///
3059 /// This intrinsic corresponds to the <c> VBROADCASTSS </c> instruction.
3060 ///
3061 /// \param __a
3062 ///    The single-precision floating point value to be broadcast.
3063 /// \returns A 256-bit vector of [8 x float] whose 32-bit elements are set
3064 ///    equal to the broadcast value.
3065 static __inline __m256 __DEFAULT_FN_ATTRS
3066 _mm256_broadcast_ss(float const *__a)
3067 {
3068   struct __mm256_broadcast_ss_struct {
3069     float __f;
3070   } __attribute__((__packed__, __may_alias__));
3071   float __f = ((const struct __mm256_broadcast_ss_struct*)__a)->__f;
3072   return __extension__ (__m256)(__v8sf){ __f, __f, __f, __f, __f, __f, __f, __f };
3073 }
3074 
3075 /// Loads the data from a 128-bit vector of [2 x double] from the
3076 ///    specified address pointed to by \a __a and broadcasts it to 128-bit
3077 ///    elements in a 256-bit vector of [4 x double].
3078 ///
3079 /// \headerfile <x86intrin.h>
3080 ///
3081 /// This intrinsic corresponds to the <c> VBROADCASTF128 </c> instruction.
3082 ///
3083 /// \param __a
3084 ///    The 128-bit vector of [2 x double] to be broadcast.
3085 /// \returns A 256-bit vector of [4 x double] whose 128-bit elements are set
3086 ///    equal to the broadcast value.
3087 static __inline __m256d __DEFAULT_FN_ATTRS
3088 _mm256_broadcast_pd(__m128d const *__a)
3089 {
3090   __m128d __b = _mm_loadu_pd((const double *)__a);
3091   return (__m256d)__builtin_shufflevector((__v2df)__b, (__v2df)__b,
3092                                           0, 1, 0, 1);
3093 }
3094 
3095 /// Loads the data from a 128-bit vector of [4 x float] from the
3096 ///    specified address pointed to by \a __a and broadcasts it to 128-bit
3097 ///    elements in a 256-bit vector of [8 x float].
3098 ///
3099 /// \headerfile <x86intrin.h>
3100 ///
3101 /// This intrinsic corresponds to the <c> VBROADCASTF128 </c> instruction.
3102 ///
3103 /// \param __a
3104 ///    The 128-bit vector of [4 x float] to be broadcast.
3105 /// \returns A 256-bit vector of [8 x float] whose 128-bit elements are set
3106 ///    equal to the broadcast value.
3107 static __inline __m256 __DEFAULT_FN_ATTRS
3108 _mm256_broadcast_ps(__m128 const *__a)
3109 {
3110   __m128 __b = _mm_loadu_ps((const float *)__a);
3111   return (__m256)__builtin_shufflevector((__v4sf)__b, (__v4sf)__b,
3112                                          0, 1, 2, 3, 0, 1, 2, 3);
3113 }
3114 
3115 /* SIMD load ops */
3116 /// Loads 4 double-precision floating point values from a 32-byte aligned
3117 ///    memory location pointed to by \a __p into a vector of [4 x double].
3118 ///
3119 /// \headerfile <x86intrin.h>
3120 ///
3121 /// This intrinsic corresponds to the <c> VMOVAPD </c> instruction.
3122 ///
3123 /// \param __p
3124 ///    A 32-byte aligned pointer to a memory location containing
3125 ///    double-precision floating point values.
3126 /// \returns A 256-bit vector of [4 x double] containing the moved values.
3127 static __inline __m256d __DEFAULT_FN_ATTRS
3128 _mm256_load_pd(double const *__p)
3129 {
3130   return *(const __m256d *)__p;
3131 }
3132 
3133 /// Loads 8 single-precision floating point values from a 32-byte aligned
3134 ///    memory location pointed to by \a __p into a vector of [8 x float].
3135 ///
3136 /// \headerfile <x86intrin.h>
3137 ///
3138 /// This intrinsic corresponds to the <c> VMOVAPS </c> instruction.
3139 ///
3140 /// \param __p
3141 ///    A 32-byte aligned pointer to a memory location containing float values.
3142 /// \returns A 256-bit vector of [8 x float] containing the moved values.
3143 static __inline __m256 __DEFAULT_FN_ATTRS
3144 _mm256_load_ps(float const *__p)
3145 {
3146   return *(const __m256 *)__p;
3147 }
3148 
3149 /// Loads 4 double-precision floating point values from an unaligned
3150 ///    memory location pointed to by \a __p into a vector of [4 x double].
3151 ///
3152 /// \headerfile <x86intrin.h>
3153 ///
3154 /// This intrinsic corresponds to the <c> VMOVUPD </c> instruction.
3155 ///
3156 /// \param __p
3157 ///    A pointer to a memory location containing double-precision floating
3158 ///    point values.
3159 /// \returns A 256-bit vector of [4 x double] containing the moved values.
3160 static __inline __m256d __DEFAULT_FN_ATTRS
3161 _mm256_loadu_pd(double const *__p)
3162 {
3163   struct __loadu_pd {
3164     __m256d_u __v;
3165   } __attribute__((__packed__, __may_alias__));
3166   return ((const struct __loadu_pd*)__p)->__v;
3167 }
3168 
3169 /// Loads 8 single-precision floating point values from an unaligned
3170 ///    memory location pointed to by \a __p into a vector of [8 x float].
3171 ///
3172 /// \headerfile <x86intrin.h>
3173 ///
3174 /// This intrinsic corresponds to the <c> VMOVUPS </c> instruction.
3175 ///
3176 /// \param __p
3177 ///    A pointer to a memory location containing single-precision floating
3178 ///    point values.
3179 /// \returns A 256-bit vector of [8 x float] containing the moved values.
3180 static __inline __m256 __DEFAULT_FN_ATTRS
3181 _mm256_loadu_ps(float const *__p)
3182 {
3183   struct __loadu_ps {
3184     __m256_u __v;
3185   } __attribute__((__packed__, __may_alias__));
3186   return ((const struct __loadu_ps*)__p)->__v;
3187 }
3188 
3189 /// Loads 256 bits of integer data from a 32-byte aligned memory
3190 ///    location pointed to by \a __p into elements of a 256-bit integer vector.
3191 ///
3192 /// \headerfile <x86intrin.h>
3193 ///
3194 /// This intrinsic corresponds to the <c> VMOVDQA </c> instruction.
3195 ///
3196 /// \param __p
3197 ///    A 32-byte aligned pointer to a 256-bit integer vector containing integer
3198 ///    values.
3199 /// \returns A 256-bit integer vector containing the moved values.
3200 static __inline __m256i __DEFAULT_FN_ATTRS
3201 _mm256_load_si256(__m256i const *__p)
3202 {
3203   return *__p;
3204 }
3205 
3206 /// Loads 256 bits of integer data from an unaligned memory location
3207 ///    pointed to by \a __p into a 256-bit integer vector.
3208 ///
3209 /// \headerfile <x86intrin.h>
3210 ///
3211 /// This intrinsic corresponds to the <c> VMOVDQU </c> instruction.
3212 ///
3213 /// \param __p
3214 ///    A pointer to a 256-bit integer vector containing integer values.
3215 /// \returns A 256-bit integer vector containing the moved values.
3216 static __inline __m256i __DEFAULT_FN_ATTRS
3217 _mm256_loadu_si256(__m256i_u const *__p)
3218 {
3219   struct __loadu_si256 {
3220     __m256i_u __v;
3221   } __attribute__((__packed__, __may_alias__));
3222   return ((const struct __loadu_si256*)__p)->__v;
3223 }
3224 
3225 /// Loads 256 bits of integer data from an unaligned memory location
3226 ///    pointed to by \a __p into a 256-bit integer vector. This intrinsic may
3227 ///    perform better than \c _mm256_loadu_si256 when the data crosses a cache
3228 ///    line boundary.
3229 ///
3230 /// \headerfile <x86intrin.h>
3231 ///
3232 /// This intrinsic corresponds to the <c> VLDDQU </c> instruction.
3233 ///
3234 /// \param __p
3235 ///    A pointer to a 256-bit integer vector containing integer values.
3236 /// \returns A 256-bit integer vector containing the moved values.
3237 static __inline __m256i __DEFAULT_FN_ATTRS
3238 _mm256_lddqu_si256(__m256i_u const *__p)
3239 {
3240   return (__m256i)__builtin_ia32_lddqu256((char const *)__p);
3241 }
3242 
3243 /* SIMD store ops */
3244 /// Stores double-precision floating point values from a 256-bit vector
3245 ///    of [4 x double] to a 32-byte aligned memory location pointed to by
3246 ///    \a __p.
3247 ///
3248 /// \headerfile <x86intrin.h>
3249 ///
3250 /// This intrinsic corresponds to the <c> VMOVAPD </c> instruction.
3251 ///
3252 /// \param __p
3253 ///    A 32-byte aligned pointer to a memory location that will receive the
3254 ///    double-precision floaing point values.
3255 /// \param __a
3256 ///    A 256-bit vector of [4 x double] containing the values to be moved.
3257 static __inline void __DEFAULT_FN_ATTRS
3258 _mm256_store_pd(double *__p, __m256d __a)
3259 {
3260   *(__m256d *)__p = __a;
3261 }
3262 
3263 /// Stores single-precision floating point values from a 256-bit vector
3264 ///    of [8 x float] to a 32-byte aligned memory location pointed to by \a __p.
3265 ///
3266 /// \headerfile <x86intrin.h>
3267 ///
3268 /// This intrinsic corresponds to the <c> VMOVAPS </c> instruction.
3269 ///
3270 /// \param __p
3271 ///    A 32-byte aligned pointer to a memory location that will receive the
3272 ///    float values.
3273 /// \param __a
3274 ///    A 256-bit vector of [8 x float] containing the values to be moved.
3275 static __inline void __DEFAULT_FN_ATTRS
3276 _mm256_store_ps(float *__p, __m256 __a)
3277 {
3278   *(__m256 *)__p = __a;
3279 }
3280 
3281 /// Stores double-precision floating point values from a 256-bit vector
3282 ///    of [4 x double] to an unaligned memory location pointed to by \a __p.
3283 ///
3284 /// \headerfile <x86intrin.h>
3285 ///
3286 /// This intrinsic corresponds to the <c> VMOVUPD </c> instruction.
3287 ///
3288 /// \param __p
3289 ///    A pointer to a memory location that will receive the double-precision
3290 ///    floating point values.
3291 /// \param __a
3292 ///    A 256-bit vector of [4 x double] containing the values to be moved.
3293 static __inline void __DEFAULT_FN_ATTRS
3294 _mm256_storeu_pd(double *__p, __m256d __a)
3295 {
3296   struct __storeu_pd {
3297     __m256d_u __v;
3298   } __attribute__((__packed__, __may_alias__));
3299   ((struct __storeu_pd*)__p)->__v = __a;
3300 }
3301 
3302 /// Stores single-precision floating point values from a 256-bit vector
3303 ///    of [8 x float] to an unaligned memory location pointed to by \a __p.
3304 ///
3305 /// \headerfile <x86intrin.h>
3306 ///
3307 /// This intrinsic corresponds to the <c> VMOVUPS </c> instruction.
3308 ///
3309 /// \param __p
3310 ///    A pointer to a memory location that will receive the float values.
3311 /// \param __a
3312 ///    A 256-bit vector of [8 x float] containing the values to be moved.
3313 static __inline void __DEFAULT_FN_ATTRS
3314 _mm256_storeu_ps(float *__p, __m256 __a)
3315 {
3316   struct __storeu_ps {
3317     __m256_u __v;
3318   } __attribute__((__packed__, __may_alias__));
3319   ((struct __storeu_ps*)__p)->__v = __a;
3320 }
3321 
3322 /// Stores integer values from a 256-bit integer vector to a 32-byte
3323 ///    aligned memory location pointed to by \a __p.
3324 ///
3325 /// \headerfile <x86intrin.h>
3326 ///
3327 /// This intrinsic corresponds to the <c> VMOVDQA </c> instruction.
3328 ///
3329 /// \param __p
3330 ///    A 32-byte aligned pointer to a memory location that will receive the
3331 ///    integer values.
3332 /// \param __a
3333 ///    A 256-bit integer vector containing the values to be moved.
3334 static __inline void __DEFAULT_FN_ATTRS
3335 _mm256_store_si256(__m256i *__p, __m256i __a)
3336 {
3337   *__p = __a;
3338 }
3339 
3340 /// Stores integer values from a 256-bit integer vector to an unaligned
3341 ///    memory location pointed to by \a __p.
3342 ///
3343 /// \headerfile <x86intrin.h>
3344 ///
3345 /// This intrinsic corresponds to the <c> VMOVDQU </c> instruction.
3346 ///
3347 /// \param __p
3348 ///    A pointer to a memory location that will receive the integer values.
3349 /// \param __a
3350 ///    A 256-bit integer vector containing the values to be moved.
3351 static __inline void __DEFAULT_FN_ATTRS
3352 _mm256_storeu_si256(__m256i_u *__p, __m256i __a)
3353 {
3354   struct __storeu_si256 {
3355     __m256i_u __v;
3356   } __attribute__((__packed__, __may_alias__));
3357   ((struct __storeu_si256*)__p)->__v = __a;
3358 }
3359 
3360 /* Conditional load ops */
3361 /// Conditionally loads double-precision floating point elements from a
3362 ///    memory location pointed to by \a __p into a 128-bit vector of
3363 ///    [2 x double], depending on the mask bits associated with each data
3364 ///    element.
3365 ///
3366 /// \headerfile <x86intrin.h>
3367 ///
3368 /// This intrinsic corresponds to the <c> VMASKMOVPD </c> instruction.
3369 ///
3370 /// \param __p
3371 ///    A pointer to a memory location that contains the double-precision
3372 ///    floating point values.
3373 /// \param __m
3374 ///    A 128-bit integer vector containing the mask. The most significant bit of
3375 ///    each data element represents the mask bits. If a mask bit is zero, the
3376 ///    corresponding value in the memory location is not loaded and the
3377 ///    corresponding field in the return value is set to zero.
3378 /// \returns A 128-bit vector of [2 x double] containing the loaded values.
3379 static __inline __m128d __DEFAULT_FN_ATTRS128
3380 _mm_maskload_pd(double const *__p, __m128i __m)
3381 {
3382   return (__m128d)__builtin_ia32_maskloadpd((const __v2df *)__p, (__v2di)__m);
3383 }
3384 
3385 /// Conditionally loads double-precision floating point elements from a
3386 ///    memory location pointed to by \a __p into a 256-bit vector of
3387 ///    [4 x double], depending on the mask bits associated with each data
3388 ///    element.
3389 ///
3390 /// \headerfile <x86intrin.h>
3391 ///
3392 /// This intrinsic corresponds to the <c> VMASKMOVPD </c> instruction.
3393 ///
3394 /// \param __p
3395 ///    A pointer to a memory location that contains the double-precision
3396 ///    floating point values.
3397 /// \param __m
3398 ///    A 256-bit integer vector of [4 x quadword] containing the mask. The most
3399 ///    significant bit of each quadword element represents the mask bits. If a
3400 ///    mask bit is zero, the corresponding value in the memory location is not
3401 ///    loaded and the corresponding field in the return value is set to zero.
3402 /// \returns A 256-bit vector of [4 x double] containing the loaded values.
3403 static __inline __m256d __DEFAULT_FN_ATTRS
3404 _mm256_maskload_pd(double const *__p, __m256i __m)
3405 {
3406   return (__m256d)__builtin_ia32_maskloadpd256((const __v4df *)__p,
3407                                                (__v4di)__m);
3408 }
3409 
3410 /// Conditionally loads single-precision floating point elements from a
3411 ///    memory location pointed to by \a __p into a 128-bit vector of
3412 ///    [4 x float], depending on the mask bits associated with each data
3413 ///    element.
3414 ///
3415 /// \headerfile <x86intrin.h>
3416 ///
3417 /// This intrinsic corresponds to the <c> VMASKMOVPS </c> instruction.
3418 ///
3419 /// \param __p
3420 ///    A pointer to a memory location that contains the single-precision
3421 ///    floating point values.
3422 /// \param __m
3423 ///    A 128-bit integer vector containing the mask. The most significant bit of
3424 ///    each data element represents the mask bits. If a mask bit is zero, the
3425 ///    corresponding value in the memory location is not loaded and the
3426 ///    corresponding field in the return value is set to zero.
3427 /// \returns A 128-bit vector of [4 x float] containing the loaded values.
3428 static __inline __m128 __DEFAULT_FN_ATTRS128
3429 _mm_maskload_ps(float const *__p, __m128i __m)
3430 {
3431   return (__m128)__builtin_ia32_maskloadps((const __v4sf *)__p, (__v4si)__m);
3432 }
3433 
3434 /// Conditionally loads single-precision floating point elements from a
3435 ///    memory location pointed to by \a __p into a 256-bit vector of
3436 ///    [8 x float], depending on the mask bits associated with each data
3437 ///    element.
3438 ///
3439 /// \headerfile <x86intrin.h>
3440 ///
3441 /// This intrinsic corresponds to the <c> VMASKMOVPS </c> instruction.
3442 ///
3443 /// \param __p
3444 ///    A pointer to a memory location that contains the single-precision
3445 ///    floating point values.
3446 /// \param __m
3447 ///    A 256-bit integer vector of [8 x dword] containing the mask. The most
3448 ///    significant bit of each dword element represents the mask bits. If a mask
3449 ///    bit is zero, the corresponding value in the memory location is not loaded
3450 ///    and the corresponding field in the return value is set to zero.
3451 /// \returns A 256-bit vector of [8 x float] containing the loaded values.
3452 static __inline __m256 __DEFAULT_FN_ATTRS
3453 _mm256_maskload_ps(float const *__p, __m256i __m)
3454 {
3455   return (__m256)__builtin_ia32_maskloadps256((const __v8sf *)__p, (__v8si)__m);
3456 }
3457 
3458 /* Conditional store ops */
3459 /// Moves single-precision floating point values from a 256-bit vector
3460 ///    of [8 x float] to a memory location pointed to by \a __p, according to
3461 ///    the specified mask.
3462 ///
3463 /// \headerfile <x86intrin.h>
3464 ///
3465 /// This intrinsic corresponds to the <c> VMASKMOVPS </c> instruction.
3466 ///
3467 /// \param __p
3468 ///    A pointer to a memory location that will receive the float values.
3469 /// \param __m
3470 ///    A 256-bit integer vector of [8 x dword] containing the mask. The most
3471 ///    significant bit of each dword element in the mask vector represents the
3472 ///    mask bits. If a mask bit is zero, the corresponding value from vector
3473 ///    \a __a is not stored and the corresponding field in the memory location
3474 ///    pointed to by \a __p is not changed.
3475 /// \param __a
3476 ///    A 256-bit vector of [8 x float] containing the values to be stored.
3477 static __inline void __DEFAULT_FN_ATTRS
3478 _mm256_maskstore_ps(float *__p, __m256i __m, __m256 __a)
3479 {
3480   __builtin_ia32_maskstoreps256((__v8sf *)__p, (__v8si)__m, (__v8sf)__a);
3481 }
3482 
3483 /// Moves double-precision values from a 128-bit vector of [2 x double]
3484 ///    to a memory location pointed to by \a __p, according to the specified
3485 ///    mask.
3486 ///
3487 /// \headerfile <x86intrin.h>
3488 ///
3489 /// This intrinsic corresponds to the <c> VMASKMOVPD </c> instruction.
3490 ///
3491 /// \param __p
3492 ///    A pointer to a memory location that will receive the float values.
3493 /// \param __m
3494 ///    A 128-bit integer vector containing the mask. The most significant bit of
3495 ///    each field in the mask vector represents the mask bits. If a mask bit is
3496 ///    zero, the corresponding value from vector \a __a is not stored and the
3497 ///    corresponding field in the memory location pointed to by \a __p is not
3498 ///    changed.
3499 /// \param __a
3500 ///    A 128-bit vector of [2 x double] containing the values to be stored.
3501 static __inline void __DEFAULT_FN_ATTRS128
3502 _mm_maskstore_pd(double *__p, __m128i __m, __m128d __a)
3503 {
3504   __builtin_ia32_maskstorepd((__v2df *)__p, (__v2di)__m, (__v2df)__a);
3505 }
3506 
3507 /// Moves double-precision values from a 256-bit vector of [4 x double]
3508 ///    to a memory location pointed to by \a __p, according to the specified
3509 ///    mask.
3510 ///
3511 /// \headerfile <x86intrin.h>
3512 ///
3513 /// This intrinsic corresponds to the <c> VMASKMOVPD </c> instruction.
3514 ///
3515 /// \param __p
3516 ///    A pointer to a memory location that will receive the float values.
3517 /// \param __m
3518 ///    A 256-bit integer vector of [4 x quadword] containing the mask. The most
3519 ///    significant bit of each quadword element in the mask vector represents
3520 ///    the mask bits. If a mask bit is zero, the corresponding value from vector
3521 ///    __a is not stored and the corresponding field in the memory location
3522 ///    pointed to by \a __p is not changed.
3523 /// \param __a
3524 ///    A 256-bit vector of [4 x double] containing the values to be stored.
3525 static __inline void __DEFAULT_FN_ATTRS
3526 _mm256_maskstore_pd(double *__p, __m256i __m, __m256d __a)
3527 {
3528   __builtin_ia32_maskstorepd256((__v4df *)__p, (__v4di)__m, (__v4df)__a);
3529 }
3530 
3531 /// Moves single-precision floating point values from a 128-bit vector
3532 ///    of [4 x float] to a memory location pointed to by \a __p, according to
3533 ///    the specified mask.
3534 ///
3535 /// \headerfile <x86intrin.h>
3536 ///
3537 /// This intrinsic corresponds to the <c> VMASKMOVPS </c> instruction.
3538 ///
3539 /// \param __p
3540 ///    A pointer to a memory location that will receive the float values.
3541 /// \param __m
3542 ///    A 128-bit integer vector containing the mask. The most significant bit of
3543 ///    each field in the mask vector represents the mask bits. If a mask bit is
3544 ///    zero, the corresponding value from vector __a is not stored and the
3545 ///    corresponding field in the memory location pointed to by \a __p is not
3546 ///    changed.
3547 /// \param __a
3548 ///    A 128-bit vector of [4 x float] containing the values to be stored.
3549 static __inline void __DEFAULT_FN_ATTRS128
3550 _mm_maskstore_ps(float *__p, __m128i __m, __m128 __a)
3551 {
3552   __builtin_ia32_maskstoreps((__v4sf *)__p, (__v4si)__m, (__v4sf)__a);
3553 }
3554 
3555 /* Cacheability support ops */
3556 /// Moves integer data from a 256-bit integer vector to a 32-byte
3557 ///    aligned memory location. To minimize caching, the data is flagged as
3558 ///    non-temporal (unlikely to be used again soon).
3559 ///
3560 /// \headerfile <x86intrin.h>
3561 ///
3562 /// This intrinsic corresponds to the <c> VMOVNTDQ </c> instruction.
3563 ///
3564 /// \param __a
3565 ///    A pointer to a 32-byte aligned memory location that will receive the
3566 ///    integer values.
3567 /// \param __b
3568 ///    A 256-bit integer vector containing the values to be moved.
3569 static __inline void __DEFAULT_FN_ATTRS
3570 _mm256_stream_si256(void *__a, __m256i __b)
3571 {
3572   typedef __v4di __v4di_aligned __attribute__((aligned(32)));
3573   __builtin_nontemporal_store((__v4di_aligned)__b, (__v4di_aligned*)__a);
3574 }
3575 
3576 /// Moves double-precision values from a 256-bit vector of [4 x double]
3577 ///    to a 32-byte aligned memory location. To minimize caching, the data is
3578 ///    flagged as non-temporal (unlikely to be used again soon).
3579 ///
3580 /// \headerfile <x86intrin.h>
3581 ///
3582 /// This intrinsic corresponds to the <c> VMOVNTPD </c> instruction.
3583 ///
3584 /// \param __a
3585 ///    A pointer to a 32-byte aligned memory location that will receive the
3586 ///    double-precision floating-point values.
3587 /// \param __b
3588 ///    A 256-bit vector of [4 x double] containing the values to be moved.
3589 static __inline void __DEFAULT_FN_ATTRS
3590 _mm256_stream_pd(void *__a, __m256d __b)
3591 {
3592   typedef __v4df __v4df_aligned __attribute__((aligned(32)));
3593   __builtin_nontemporal_store((__v4df_aligned)__b, (__v4df_aligned*)__a);
3594 }
3595 
3596 /// Moves single-precision floating point values from a 256-bit vector
3597 ///    of [8 x float] to a 32-byte aligned memory location. To minimize
3598 ///    caching, the data is flagged as non-temporal (unlikely to be used again
3599 ///    soon).
3600 ///
3601 /// \headerfile <x86intrin.h>
3602 ///
3603 /// This intrinsic corresponds to the <c> VMOVNTPS </c> instruction.
3604 ///
3605 /// \param __p
3606 ///    A pointer to a 32-byte aligned memory location that will receive the
3607 ///    single-precision floating point values.
3608 /// \param __a
3609 ///    A 256-bit vector of [8 x float] containing the values to be moved.
3610 static __inline void __DEFAULT_FN_ATTRS
3611 _mm256_stream_ps(void *__p, __m256 __a)
3612 {
3613   typedef __v8sf __v8sf_aligned __attribute__((aligned(32)));
3614   __builtin_nontemporal_store((__v8sf_aligned)__a, (__v8sf_aligned*)__p);
3615 }
3616 
3617 /* Create vectors */
3618 /// Create a 256-bit vector of [4 x double] with undefined values.
3619 ///
3620 /// \headerfile <x86intrin.h>
3621 ///
3622 /// This intrinsic has no corresponding instruction.
3623 ///
3624 /// \returns A 256-bit vector of [4 x double] containing undefined values.
3625 static __inline__ __m256d __DEFAULT_FN_ATTRS
3626 _mm256_undefined_pd(void)
3627 {
3628   return (__m256d)__builtin_ia32_undef256();
3629 }
3630 
3631 /// Create a 256-bit vector of [8 x float] with undefined values.
3632 ///
3633 /// \headerfile <x86intrin.h>
3634 ///
3635 /// This intrinsic has no corresponding instruction.
3636 ///
3637 /// \returns A 256-bit vector of [8 x float] containing undefined values.
3638 static __inline__ __m256 __DEFAULT_FN_ATTRS
3639 _mm256_undefined_ps(void)
3640 {
3641   return (__m256)__builtin_ia32_undef256();
3642 }
3643 
3644 /// Create a 256-bit integer vector with undefined values.
3645 ///
3646 /// \headerfile <x86intrin.h>
3647 ///
3648 /// This intrinsic has no corresponding instruction.
3649 ///
3650 /// \returns A 256-bit integer vector containing undefined values.
3651 static __inline__ __m256i __DEFAULT_FN_ATTRS
3652 _mm256_undefined_si256(void)
3653 {
3654   return (__m256i)__builtin_ia32_undef256();
3655 }
3656 
3657 /// Constructs a 256-bit floating-point vector of [4 x double]
3658 ///    initialized with the specified double-precision floating-point values.
3659 ///
3660 /// \headerfile <x86intrin.h>
3661 ///
3662 /// This intrinsic corresponds to the <c> VUNPCKLPD+VINSERTF128 </c>
3663 ///   instruction.
3664 ///
3665 /// \param __a
3666 ///    A double-precision floating-point value used to initialize bits [255:192]
3667 ///    of the result.
3668 /// \param __b
3669 ///    A double-precision floating-point value used to initialize bits [191:128]
3670 ///    of the result.
3671 /// \param __c
3672 ///    A double-precision floating-point value used to initialize bits [127:64]
3673 ///    of the result.
3674 /// \param __d
3675 ///    A double-precision floating-point value used to initialize bits [63:0]
3676 ///    of the result.
3677 /// \returns An initialized 256-bit floating-point vector of [4 x double].
3678 static __inline __m256d __DEFAULT_FN_ATTRS
3679 _mm256_set_pd(double __a, double __b, double __c, double __d)
3680 {
3681   return __extension__ (__m256d){ __d, __c, __b, __a };
3682 }
3683 
3684 /// Constructs a 256-bit floating-point vector of [8 x float] initialized
3685 ///    with the specified single-precision floating-point values.
3686 ///
3687 /// \headerfile <x86intrin.h>
3688 ///
3689 /// This intrinsic is a utility function and does not correspond to a specific
3690 ///   instruction.
3691 ///
3692 /// \param __a
3693 ///    A single-precision floating-point value used to initialize bits [255:224]
3694 ///    of the result.
3695 /// \param __b
3696 ///    A single-precision floating-point value used to initialize bits [223:192]
3697 ///    of the result.
3698 /// \param __c
3699 ///    A single-precision floating-point value used to initialize bits [191:160]
3700 ///    of the result.
3701 /// \param __d
3702 ///    A single-precision floating-point value used to initialize bits [159:128]
3703 ///    of the result.
3704 /// \param __e
3705 ///    A single-precision floating-point value used to initialize bits [127:96]
3706 ///    of the result.
3707 /// \param __f
3708 ///    A single-precision floating-point value used to initialize bits [95:64]
3709 ///    of the result.
3710 /// \param __g
3711 ///    A single-precision floating-point value used to initialize bits [63:32]
3712 ///    of the result.
3713 /// \param __h
3714 ///    A single-precision floating-point value used to initialize bits [31:0]
3715 ///    of the result.
3716 /// \returns An initialized 256-bit floating-point vector of [8 x float].
3717 static __inline __m256 __DEFAULT_FN_ATTRS
3718 _mm256_set_ps(float __a, float __b, float __c, float __d,
3719               float __e, float __f, float __g, float __h)
3720 {
3721   return __extension__ (__m256){ __h, __g, __f, __e, __d, __c, __b, __a };
3722 }
3723 
3724 /// Constructs a 256-bit integer vector initialized with the specified
3725 ///    32-bit integral values.
3726 ///
3727 /// \headerfile <x86intrin.h>
3728 ///
3729 /// This intrinsic is a utility function and does not correspond to a specific
3730 ///   instruction.
3731 ///
3732 /// \param __i0
3733 ///    A 32-bit integral value used to initialize bits [255:224] of the result.
3734 /// \param __i1
3735 ///    A 32-bit integral value used to initialize bits [223:192] of the result.
3736 /// \param __i2
3737 ///    A 32-bit integral value used to initialize bits [191:160] of the result.
3738 /// \param __i3
3739 ///    A 32-bit integral value used to initialize bits [159:128] of the result.
3740 /// \param __i4
3741 ///    A 32-bit integral value used to initialize bits [127:96] of the result.
3742 /// \param __i5
3743 ///    A 32-bit integral value used to initialize bits [95:64] of the result.
3744 /// \param __i6
3745 ///    A 32-bit integral value used to initialize bits [63:32] of the result.
3746 /// \param __i7
3747 ///    A 32-bit integral value used to initialize bits [31:0] of the result.
3748 /// \returns An initialized 256-bit integer vector.
3749 static __inline __m256i __DEFAULT_FN_ATTRS
3750 _mm256_set_epi32(int __i0, int __i1, int __i2, int __i3,
3751                  int __i4, int __i5, int __i6, int __i7)
3752 {
3753   return __extension__ (__m256i)(__v8si){ __i7, __i6, __i5, __i4, __i3, __i2, __i1, __i0 };
3754 }
3755 
3756 /// Constructs a 256-bit integer vector initialized with the specified
3757 ///    16-bit integral values.
3758 ///
3759 /// \headerfile <x86intrin.h>
3760 ///
3761 /// This intrinsic is a utility function and does not correspond to a specific
3762 ///   instruction.
3763 ///
3764 /// \param __w15
3765 ///    A 16-bit integral value used to initialize bits [255:240] of the result.
3766 /// \param __w14
3767 ///    A 16-bit integral value used to initialize bits [239:224] of the result.
3768 /// \param __w13
3769 ///    A 16-bit integral value used to initialize bits [223:208] of the result.
3770 /// \param __w12
3771 ///    A 16-bit integral value used to initialize bits [207:192] of the result.
3772 /// \param __w11
3773 ///    A 16-bit integral value used to initialize bits [191:176] of the result.
3774 /// \param __w10
3775 ///    A 16-bit integral value used to initialize bits [175:160] of the result.
3776 /// \param __w09
3777 ///    A 16-bit integral value used to initialize bits [159:144] of the result.
3778 /// \param __w08
3779 ///    A 16-bit integral value used to initialize bits [143:128] of the result.
3780 /// \param __w07
3781 ///    A 16-bit integral value used to initialize bits [127:112] of the result.
3782 /// \param __w06
3783 ///    A 16-bit integral value used to initialize bits [111:96] of the result.
3784 /// \param __w05
3785 ///    A 16-bit integral value used to initialize bits [95:80] of the result.
3786 /// \param __w04
3787 ///    A 16-bit integral value used to initialize bits [79:64] of the result.
3788 /// \param __w03
3789 ///    A 16-bit integral value used to initialize bits [63:48] of the result.
3790 /// \param __w02
3791 ///    A 16-bit integral value used to initialize bits [47:32] of the result.
3792 /// \param __w01
3793 ///    A 16-bit integral value used to initialize bits [31:16] of the result.
3794 /// \param __w00
3795 ///    A 16-bit integral value used to initialize bits [15:0] of the result.
3796 /// \returns An initialized 256-bit integer vector.
3797 static __inline __m256i __DEFAULT_FN_ATTRS
3798 _mm256_set_epi16(short __w15, short __w14, short __w13, short __w12,
3799                  short __w11, short __w10, short __w09, short __w08,
3800                  short __w07, short __w06, short __w05, short __w04,
3801                  short __w03, short __w02, short __w01, short __w00)
3802 {
3803   return __extension__ (__m256i)(__v16hi){ __w00, __w01, __w02, __w03, __w04, __w05, __w06,
3804     __w07, __w08, __w09, __w10, __w11, __w12, __w13, __w14, __w15 };
3805 }
3806 
3807 /// Constructs a 256-bit integer vector initialized with the specified
3808 ///    8-bit integral values.
3809 ///
3810 /// \headerfile <x86intrin.h>
3811 ///
3812 /// This intrinsic is a utility function and does not correspond to a specific
3813 ///   instruction.
3814 ///
3815 /// \param __b31
3816 ///    An 8-bit integral value used to initialize bits [255:248] of the result.
3817 /// \param __b30
3818 ///    An 8-bit integral value used to initialize bits [247:240] of the result.
3819 /// \param __b29
3820 ///    An 8-bit integral value used to initialize bits [239:232] of the result.
3821 /// \param __b28
3822 ///    An 8-bit integral value used to initialize bits [231:224] of the result.
3823 /// \param __b27
3824 ///    An 8-bit integral value used to initialize bits [223:216] of the result.
3825 /// \param __b26
3826 ///    An 8-bit integral value used to initialize bits [215:208] of the result.
3827 /// \param __b25
3828 ///    An 8-bit integral value used to initialize bits [207:200] of the result.
3829 /// \param __b24
3830 ///    An 8-bit integral value used to initialize bits [199:192] of the result.
3831 /// \param __b23
3832 ///    An 8-bit integral value used to initialize bits [191:184] of the result.
3833 /// \param __b22
3834 ///    An 8-bit integral value used to initialize bits [183:176] of the result.
3835 /// \param __b21
3836 ///    An 8-bit integral value used to initialize bits [175:168] of the result.
3837 /// \param __b20
3838 ///    An 8-bit integral value used to initialize bits [167:160] of the result.
3839 /// \param __b19
3840 ///    An 8-bit integral value used to initialize bits [159:152] of the result.
3841 /// \param __b18
3842 ///    An 8-bit integral value used to initialize bits [151:144] of the result.
3843 /// \param __b17
3844 ///    An 8-bit integral value used to initialize bits [143:136] of the result.
3845 /// \param __b16
3846 ///    An 8-bit integral value used to initialize bits [135:128] of the result.
3847 /// \param __b15
3848 ///    An 8-bit integral value used to initialize bits [127:120] of the result.
3849 /// \param __b14
3850 ///    An 8-bit integral value used to initialize bits [119:112] of the result.
3851 /// \param __b13
3852 ///    An 8-bit integral value used to initialize bits [111:104] of the result.
3853 /// \param __b12
3854 ///    An 8-bit integral value used to initialize bits [103:96] of the result.
3855 /// \param __b11
3856 ///    An 8-bit integral value used to initialize bits [95:88] of the result.
3857 /// \param __b10
3858 ///    An 8-bit integral value used to initialize bits [87:80] of the result.
3859 /// \param __b09
3860 ///    An 8-bit integral value used to initialize bits [79:72] of the result.
3861 /// \param __b08
3862 ///    An 8-bit integral value used to initialize bits [71:64] of the result.
3863 /// \param __b07
3864 ///    An 8-bit integral value used to initialize bits [63:56] of the result.
3865 /// \param __b06
3866 ///    An 8-bit integral value used to initialize bits [55:48] of the result.
3867 /// \param __b05
3868 ///    An 8-bit integral value used to initialize bits [47:40] of the result.
3869 /// \param __b04
3870 ///    An 8-bit integral value used to initialize bits [39:32] of the result.
3871 /// \param __b03
3872 ///    An 8-bit integral value used to initialize bits [31:24] of the result.
3873 /// \param __b02
3874 ///    An 8-bit integral value used to initialize bits [23:16] of the result.
3875 /// \param __b01
3876 ///    An 8-bit integral value used to initialize bits [15:8] of the result.
3877 /// \param __b00
3878 ///    An 8-bit integral value used to initialize bits [7:0] of the result.
3879 /// \returns An initialized 256-bit integer vector.
3880 static __inline __m256i __DEFAULT_FN_ATTRS
3881 _mm256_set_epi8(char __b31, char __b30, char __b29, char __b28,
3882                 char __b27, char __b26, char __b25, char __b24,
3883                 char __b23, char __b22, char __b21, char __b20,
3884                 char __b19, char __b18, char __b17, char __b16,
3885                 char __b15, char __b14, char __b13, char __b12,
3886                 char __b11, char __b10, char __b09, char __b08,
3887                 char __b07, char __b06, char __b05, char __b04,
3888                 char __b03, char __b02, char __b01, char __b00)
3889 {
3890   return __extension__ (__m256i)(__v32qi){
3891     __b00, __b01, __b02, __b03, __b04, __b05, __b06, __b07,
3892     __b08, __b09, __b10, __b11, __b12, __b13, __b14, __b15,
3893     __b16, __b17, __b18, __b19, __b20, __b21, __b22, __b23,
3894     __b24, __b25, __b26, __b27, __b28, __b29, __b30, __b31
3895   };
3896 }
3897 
3898 /// Constructs a 256-bit integer vector initialized with the specified
3899 ///    64-bit integral values.
3900 ///
3901 /// \headerfile <x86intrin.h>
3902 ///
3903 /// This intrinsic corresponds to the <c> VPUNPCKLQDQ+VINSERTF128 </c>
3904 ///   instruction.
3905 ///
3906 /// \param __a
3907 ///    A 64-bit integral value used to initialize bits [255:192] of the result.
3908 /// \param __b
3909 ///    A 64-bit integral value used to initialize bits [191:128] of the result.
3910 /// \param __c
3911 ///    A 64-bit integral value used to initialize bits [127:64] of the result.
3912 /// \param __d
3913 ///    A 64-bit integral value used to initialize bits [63:0] of the result.
3914 /// \returns An initialized 256-bit integer vector.
3915 static __inline __m256i __DEFAULT_FN_ATTRS
3916 _mm256_set_epi64x(long long __a, long long __b, long long __c, long long __d)
3917 {
3918   return __extension__ (__m256i)(__v4di){ __d, __c, __b, __a };
3919 }
3920 
3921 /* Create vectors with elements in reverse order */
3922 /// Constructs a 256-bit floating-point vector of [4 x double],
3923 ///    initialized in reverse order with the specified double-precision
3924 ///    floating-point values.
3925 ///
3926 /// \headerfile <x86intrin.h>
3927 ///
3928 /// This intrinsic corresponds to the <c> VUNPCKLPD+VINSERTF128 </c>
3929 ///   instruction.
3930 ///
3931 /// \param __a
3932 ///    A double-precision floating-point value used to initialize bits [63:0]
3933 ///    of the result.
3934 /// \param __b
3935 ///    A double-precision floating-point value used to initialize bits [127:64]
3936 ///    of the result.
3937 /// \param __c
3938 ///    A double-precision floating-point value used to initialize bits [191:128]
3939 ///    of the result.
3940 /// \param __d
3941 ///    A double-precision floating-point value used to initialize bits [255:192]
3942 ///    of the result.
3943 /// \returns An initialized 256-bit floating-point vector of [4 x double].
3944 static __inline __m256d __DEFAULT_FN_ATTRS
3945 _mm256_setr_pd(double __a, double __b, double __c, double __d)
3946 {
3947   return _mm256_set_pd(__d, __c, __b, __a);
3948 }
3949 
3950 /// Constructs a 256-bit floating-point vector of [8 x float],
3951 ///    initialized in reverse order with the specified single-precision
3952 ///    float-point values.
3953 ///
3954 /// \headerfile <x86intrin.h>
3955 ///
3956 /// This intrinsic is a utility function and does not correspond to a specific
3957 ///   instruction.
3958 ///
3959 /// \param __a
3960 ///    A single-precision floating-point value used to initialize bits [31:0]
3961 ///    of the result.
3962 /// \param __b
3963 ///    A single-precision floating-point value used to initialize bits [63:32]
3964 ///    of the result.
3965 /// \param __c
3966 ///    A single-precision floating-point value used to initialize bits [95:64]
3967 ///    of the result.
3968 /// \param __d
3969 ///    A single-precision floating-point value used to initialize bits [127:96]
3970 ///    of the result.
3971 /// \param __e
3972 ///    A single-precision floating-point value used to initialize bits [159:128]
3973 ///    of the result.
3974 /// \param __f
3975 ///    A single-precision floating-point value used to initialize bits [191:160]
3976 ///    of the result.
3977 /// \param __g
3978 ///    A single-precision floating-point value used to initialize bits [223:192]
3979 ///    of the result.
3980 /// \param __h
3981 ///    A single-precision floating-point value used to initialize bits [255:224]
3982 ///    of the result.
3983 /// \returns An initialized 256-bit floating-point vector of [8 x float].
3984 static __inline __m256 __DEFAULT_FN_ATTRS
3985 _mm256_setr_ps(float __a, float __b, float __c, float __d,
3986                float __e, float __f, float __g, float __h)
3987 {
3988   return _mm256_set_ps(__h, __g, __f, __e, __d, __c, __b, __a);
3989 }
3990 
3991 /// Constructs a 256-bit integer vector, initialized in reverse order
3992 ///    with the specified 32-bit integral values.
3993 ///
3994 /// \headerfile <x86intrin.h>
3995 ///
3996 /// This intrinsic is a utility function and does not correspond to a specific
3997 ///   instruction.
3998 ///
3999 /// \param __i0
4000 ///    A 32-bit integral value used to initialize bits [31:0] of the result.
4001 /// \param __i1
4002 ///    A 32-bit integral value used to initialize bits [63:32] of the result.
4003 /// \param __i2
4004 ///    A 32-bit integral value used to initialize bits [95:64] of the result.
4005 /// \param __i3
4006 ///    A 32-bit integral value used to initialize bits [127:96] of the result.
4007 /// \param __i4
4008 ///    A 32-bit integral value used to initialize bits [159:128] of the result.
4009 /// \param __i5
4010 ///    A 32-bit integral value used to initialize bits [191:160] of the result.
4011 /// \param __i6
4012 ///    A 32-bit integral value used to initialize bits [223:192] of the result.
4013 /// \param __i7
4014 ///    A 32-bit integral value used to initialize bits [255:224] of the result.
4015 /// \returns An initialized 256-bit integer vector.
4016 static __inline __m256i __DEFAULT_FN_ATTRS
4017 _mm256_setr_epi32(int __i0, int __i1, int __i2, int __i3,
4018                   int __i4, int __i5, int __i6, int __i7)
4019 {
4020   return _mm256_set_epi32(__i7, __i6, __i5, __i4, __i3, __i2, __i1, __i0);
4021 }
4022 
4023 /// Constructs a 256-bit integer vector, initialized in reverse order
4024 ///    with the specified 16-bit integral values.
4025 ///
4026 /// \headerfile <x86intrin.h>
4027 ///
4028 /// This intrinsic is a utility function and does not correspond to a specific
4029 ///   instruction.
4030 ///
4031 /// \param __w15
4032 ///    A 16-bit integral value used to initialize bits [15:0] of the result.
4033 /// \param __w14
4034 ///    A 16-bit integral value used to initialize bits [31:16] of the result.
4035 /// \param __w13
4036 ///    A 16-bit integral value used to initialize bits [47:32] of the result.
4037 /// \param __w12
4038 ///    A 16-bit integral value used to initialize bits [63:48] of the result.
4039 /// \param __w11
4040 ///    A 16-bit integral value used to initialize bits [79:64] of the result.
4041 /// \param __w10
4042 ///    A 16-bit integral value used to initialize bits [95:80] of the result.
4043 /// \param __w09
4044 ///    A 16-bit integral value used to initialize bits [111:96] of the result.
4045 /// \param __w08
4046 ///    A 16-bit integral value used to initialize bits [127:112] of the result.
4047 /// \param __w07
4048 ///    A 16-bit integral value used to initialize bits [143:128] of the result.
4049 /// \param __w06
4050 ///    A 16-bit integral value used to initialize bits [159:144] of the result.
4051 /// \param __w05
4052 ///    A 16-bit integral value used to initialize bits [175:160] of the result.
4053 /// \param __w04
4054 ///    A 16-bit integral value used to initialize bits [191:176] of the result.
4055 /// \param __w03
4056 ///    A 16-bit integral value used to initialize bits [207:192] of the result.
4057 /// \param __w02
4058 ///    A 16-bit integral value used to initialize bits [223:208] of the result.
4059 /// \param __w01
4060 ///    A 16-bit integral value used to initialize bits [239:224] of the result.
4061 /// \param __w00
4062 ///    A 16-bit integral value used to initialize bits [255:240] of the result.
4063 /// \returns An initialized 256-bit integer vector.
4064 static __inline __m256i __DEFAULT_FN_ATTRS
4065 _mm256_setr_epi16(short __w15, short __w14, short __w13, short __w12,
4066        short __w11, short __w10, short __w09, short __w08,
4067        short __w07, short __w06, short __w05, short __w04,
4068        short __w03, short __w02, short __w01, short __w00)
4069 {
4070   return _mm256_set_epi16(__w00, __w01, __w02, __w03,
4071                           __w04, __w05, __w06, __w07,
4072                           __w08, __w09, __w10, __w11,
4073                           __w12, __w13, __w14, __w15);
4074 }
4075 
4076 /// Constructs a 256-bit integer vector, initialized in reverse order
4077 ///    with the specified 8-bit integral values.
4078 ///
4079 /// \headerfile <x86intrin.h>
4080 ///
4081 /// This intrinsic is a utility function and does not correspond to a specific
4082 ///   instruction.
4083 ///
4084 /// \param __b31
4085 ///    An 8-bit integral value used to initialize bits [7:0] of the result.
4086 /// \param __b30
4087 ///    An 8-bit integral value used to initialize bits [15:8] of the result.
4088 /// \param __b29
4089 ///    An 8-bit integral value used to initialize bits [23:16] of the result.
4090 /// \param __b28
4091 ///    An 8-bit integral value used to initialize bits [31:24] of the result.
4092 /// \param __b27
4093 ///    An 8-bit integral value used to initialize bits [39:32] of the result.
4094 /// \param __b26
4095 ///    An 8-bit integral value used to initialize bits [47:40] of the result.
4096 /// \param __b25
4097 ///    An 8-bit integral value used to initialize bits [55:48] of the result.
4098 /// \param __b24
4099 ///    An 8-bit integral value used to initialize bits [63:56] of the result.
4100 /// \param __b23
4101 ///    An 8-bit integral value used to initialize bits [71:64] of the result.
4102 /// \param __b22
4103 ///    An 8-bit integral value used to initialize bits [79:72] of the result.
4104 /// \param __b21
4105 ///    An 8-bit integral value used to initialize bits [87:80] of the result.
4106 /// \param __b20
4107 ///    An 8-bit integral value used to initialize bits [95:88] of the result.
4108 /// \param __b19
4109 ///    An 8-bit integral value used to initialize bits [103:96] of the result.
4110 /// \param __b18
4111 ///    An 8-bit integral value used to initialize bits [111:104] of the result.
4112 /// \param __b17
4113 ///    An 8-bit integral value used to initialize bits [119:112] of the result.
4114 /// \param __b16
4115 ///    An 8-bit integral value used to initialize bits [127:120] of the result.
4116 /// \param __b15
4117 ///    An 8-bit integral value used to initialize bits [135:128] of the result.
4118 /// \param __b14
4119 ///    An 8-bit integral value used to initialize bits [143:136] of the result.
4120 /// \param __b13
4121 ///    An 8-bit integral value used to initialize bits [151:144] of the result.
4122 /// \param __b12
4123 ///    An 8-bit integral value used to initialize bits [159:152] of the result.
4124 /// \param __b11
4125 ///    An 8-bit integral value used to initialize bits [167:160] of the result.
4126 /// \param __b10
4127 ///    An 8-bit integral value used to initialize bits [175:168] of the result.
4128 /// \param __b09
4129 ///    An 8-bit integral value used to initialize bits [183:176] of the result.
4130 /// \param __b08
4131 ///    An 8-bit integral value used to initialize bits [191:184] of the result.
4132 /// \param __b07
4133 ///    An 8-bit integral value used to initialize bits [199:192] of the result.
4134 /// \param __b06
4135 ///    An 8-bit integral value used to initialize bits [207:200] of the result.
4136 /// \param __b05
4137 ///    An 8-bit integral value used to initialize bits [215:208] of the result.
4138 /// \param __b04
4139 ///    An 8-bit integral value used to initialize bits [223:216] of the result.
4140 /// \param __b03
4141 ///    An 8-bit integral value used to initialize bits [231:224] of the result.
4142 /// \param __b02
4143 ///    An 8-bit integral value used to initialize bits [239:232] of the result.
4144 /// \param __b01
4145 ///    An 8-bit integral value used to initialize bits [247:240] of the result.
4146 /// \param __b00
4147 ///    An 8-bit integral value used to initialize bits [255:248] of the result.
4148 /// \returns An initialized 256-bit integer vector.
4149 static __inline __m256i __DEFAULT_FN_ATTRS
4150 _mm256_setr_epi8(char __b31, char __b30, char __b29, char __b28,
4151                  char __b27, char __b26, char __b25, char __b24,
4152                  char __b23, char __b22, char __b21, char __b20,
4153                  char __b19, char __b18, char __b17, char __b16,
4154                  char __b15, char __b14, char __b13, char __b12,
4155                  char __b11, char __b10, char __b09, char __b08,
4156                  char __b07, char __b06, char __b05, char __b04,
4157                  char __b03, char __b02, char __b01, char __b00)
4158 {
4159   return _mm256_set_epi8(__b00, __b01, __b02, __b03, __b04, __b05, __b06, __b07,
4160                          __b08, __b09, __b10, __b11, __b12, __b13, __b14, __b15,
4161                          __b16, __b17, __b18, __b19, __b20, __b21, __b22, __b23,
4162                          __b24, __b25, __b26, __b27, __b28, __b29, __b30, __b31);
4163 }
4164 
4165 /// Constructs a 256-bit integer vector, initialized in reverse order
4166 ///    with the specified 64-bit integral values.
4167 ///
4168 /// \headerfile <x86intrin.h>
4169 ///
4170 /// This intrinsic corresponds to the <c> VPUNPCKLQDQ+VINSERTF128 </c>
4171 ///   instruction.
4172 ///
4173 /// \param __a
4174 ///    A 64-bit integral value used to initialize bits [63:0] of the result.
4175 /// \param __b
4176 ///    A 64-bit integral value used to initialize bits [127:64] of the result.
4177 /// \param __c
4178 ///    A 64-bit integral value used to initialize bits [191:128] of the result.
4179 /// \param __d
4180 ///    A 64-bit integral value used to initialize bits [255:192] of the result.
4181 /// \returns An initialized 256-bit integer vector.
4182 static __inline __m256i __DEFAULT_FN_ATTRS
4183 _mm256_setr_epi64x(long long __a, long long __b, long long __c, long long __d)
4184 {
4185   return _mm256_set_epi64x(__d, __c, __b, __a);
4186 }
4187 
4188 /* Create vectors with repeated elements */
4189 /// Constructs a 256-bit floating-point vector of [4 x double], with each
4190 ///    of the four double-precision floating-point vector elements set to the
4191 ///    specified double-precision floating-point value.
4192 ///
4193 /// \headerfile <x86intrin.h>
4194 ///
4195 /// This intrinsic corresponds to the <c> VMOVDDUP+VINSERTF128 </c> instruction.
4196 ///
4197 /// \param __w
4198 ///    A double-precision floating-point value used to initialize each vector
4199 ///    element of the result.
4200 /// \returns An initialized 256-bit floating-point vector of [4 x double].
4201 static __inline __m256d __DEFAULT_FN_ATTRS
4202 _mm256_set1_pd(double __w)
4203 {
4204   return _mm256_set_pd(__w, __w, __w, __w);
4205 }
4206 
4207 /// Constructs a 256-bit floating-point vector of [8 x float], with each
4208 ///    of the eight single-precision floating-point vector elements set to the
4209 ///    specified single-precision floating-point value.
4210 ///
4211 /// \headerfile <x86intrin.h>
4212 ///
4213 /// This intrinsic corresponds to the <c> VPERMILPS+VINSERTF128 </c>
4214 ///   instruction.
4215 ///
4216 /// \param __w
4217 ///    A single-precision floating-point value used to initialize each vector
4218 ///    element of the result.
4219 /// \returns An initialized 256-bit floating-point vector of [8 x float].
4220 static __inline __m256 __DEFAULT_FN_ATTRS
4221 _mm256_set1_ps(float __w)
4222 {
4223   return _mm256_set_ps(__w, __w, __w, __w, __w, __w, __w, __w);
4224 }
4225 
4226 /// Constructs a 256-bit integer vector of [8 x i32], with each of the
4227 ///    32-bit integral vector elements set to the specified 32-bit integral
4228 ///    value.
4229 ///
4230 /// \headerfile <x86intrin.h>
4231 ///
4232 /// This intrinsic corresponds to the <c> VPERMILPS+VINSERTF128 </c>
4233 ///   instruction.
4234 ///
4235 /// \param __i
4236 ///    A 32-bit integral value used to initialize each vector element of the
4237 ///    result.
4238 /// \returns An initialized 256-bit integer vector of [8 x i32].
4239 static __inline __m256i __DEFAULT_FN_ATTRS
4240 _mm256_set1_epi32(int __i)
4241 {
4242   return _mm256_set_epi32(__i, __i, __i, __i, __i, __i, __i, __i);
4243 }
4244 
4245 /// Constructs a 256-bit integer vector of [16 x i16], with each of the
4246 ///    16-bit integral vector elements set to the specified 16-bit integral
4247 ///    value.
4248 ///
4249 /// \headerfile <x86intrin.h>
4250 ///
4251 /// This intrinsic corresponds to the <c> VPSHUFB+VINSERTF128 </c> instruction.
4252 ///
4253 /// \param __w
4254 ///    A 16-bit integral value used to initialize each vector element of the
4255 ///    result.
4256 /// \returns An initialized 256-bit integer vector of [16 x i16].
4257 static __inline __m256i __DEFAULT_FN_ATTRS
4258 _mm256_set1_epi16(short __w)
4259 {
4260   return _mm256_set_epi16(__w, __w, __w, __w, __w, __w, __w, __w,
4261                           __w, __w, __w, __w, __w, __w, __w, __w);
4262 }
4263 
4264 /// Constructs a 256-bit integer vector of [32 x i8], with each of the
4265 ///    8-bit integral vector elements set to the specified 8-bit integral value.
4266 ///
4267 /// \headerfile <x86intrin.h>
4268 ///
4269 /// This intrinsic corresponds to the <c> VPSHUFB+VINSERTF128 </c> instruction.
4270 ///
4271 /// \param __b
4272 ///    An 8-bit integral value used to initialize each vector element of the
4273 ///    result.
4274 /// \returns An initialized 256-bit integer vector of [32 x i8].
4275 static __inline __m256i __DEFAULT_FN_ATTRS
4276 _mm256_set1_epi8(char __b)
4277 {
4278   return _mm256_set_epi8(__b, __b, __b, __b, __b, __b, __b, __b,
4279                          __b, __b, __b, __b, __b, __b, __b, __b,
4280                          __b, __b, __b, __b, __b, __b, __b, __b,
4281                          __b, __b, __b, __b, __b, __b, __b, __b);
4282 }
4283 
4284 /// Constructs a 256-bit integer vector of [4 x i64], with each of the
4285 ///    64-bit integral vector elements set to the specified 64-bit integral
4286 ///    value.
4287 ///
4288 /// \headerfile <x86intrin.h>
4289 ///
4290 /// This intrinsic corresponds to the <c> VMOVDDUP+VINSERTF128 </c> instruction.
4291 ///
4292 /// \param __q
4293 ///    A 64-bit integral value used to initialize each vector element of the
4294 ///    result.
4295 /// \returns An initialized 256-bit integer vector of [4 x i64].
4296 static __inline __m256i __DEFAULT_FN_ATTRS
4297 _mm256_set1_epi64x(long long __q)
4298 {
4299   return _mm256_set_epi64x(__q, __q, __q, __q);
4300 }
4301 
4302 /* Create __zeroed vectors */
4303 /// Constructs a 256-bit floating-point vector of [4 x double] with all
4304 ///    vector elements initialized to zero.
4305 ///
4306 /// \headerfile <x86intrin.h>
4307 ///
4308 /// This intrinsic corresponds to the <c> VXORPS </c> instruction.
4309 ///
4310 /// \returns A 256-bit vector of [4 x double] with all elements set to zero.
4311 static __inline __m256d __DEFAULT_FN_ATTRS
4312 _mm256_setzero_pd(void)
4313 {
4314   return __extension__ (__m256d){ 0.0, 0.0, 0.0, 0.0 };
4315 }
4316 
4317 /// Constructs a 256-bit floating-point vector of [8 x float] with all
4318 ///    vector elements initialized to zero.
4319 ///
4320 /// \headerfile <x86intrin.h>
4321 ///
4322 /// This intrinsic corresponds to the <c> VXORPS </c> instruction.
4323 ///
4324 /// \returns A 256-bit vector of [8 x float] with all elements set to zero.
4325 static __inline __m256 __DEFAULT_FN_ATTRS
4326 _mm256_setzero_ps(void)
4327 {
4328   return __extension__ (__m256){ 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f };
4329 }
4330 
4331 /// Constructs a 256-bit integer vector initialized to zero.
4332 ///
4333 /// \headerfile <x86intrin.h>
4334 ///
4335 /// This intrinsic corresponds to the <c> VXORPS </c> instruction.
4336 ///
4337 /// \returns A 256-bit integer vector initialized to zero.
4338 static __inline __m256i __DEFAULT_FN_ATTRS
4339 _mm256_setzero_si256(void)
4340 {
4341   return __extension__ (__m256i)(__v4di){ 0, 0, 0, 0 };
4342 }
4343 
4344 /* Cast between vector types */
4345 /// Casts a 256-bit floating-point vector of [4 x double] into a 256-bit
4346 ///    floating-point vector of [8 x float].
4347 ///
4348 /// \headerfile <x86intrin.h>
4349 ///
4350 /// This intrinsic has no corresponding instruction.
4351 ///
4352 /// \param __a
4353 ///    A 256-bit floating-point vector of [4 x double].
4354 /// \returns A 256-bit floating-point vector of [8 x float] containing the same
4355 ///    bitwise pattern as the parameter.
4356 static __inline __m256 __DEFAULT_FN_ATTRS
4357 _mm256_castpd_ps(__m256d __a)
4358 {
4359   return (__m256)__a;
4360 }
4361 
4362 /// Casts a 256-bit floating-point vector of [4 x double] into a 256-bit
4363 ///    integer vector.
4364 ///
4365 /// \headerfile <x86intrin.h>
4366 ///
4367 /// This intrinsic has no corresponding instruction.
4368 ///
4369 /// \param __a
4370 ///    A 256-bit floating-point vector of [4 x double].
4371 /// \returns A 256-bit integer vector containing the same bitwise pattern as the
4372 ///    parameter.
4373 static __inline __m256i __DEFAULT_FN_ATTRS
4374 _mm256_castpd_si256(__m256d __a)
4375 {
4376   return (__m256i)__a;
4377 }
4378 
4379 /// Casts a 256-bit floating-point vector of [8 x float] into a 256-bit
4380 ///    floating-point vector of [4 x double].
4381 ///
4382 /// \headerfile <x86intrin.h>
4383 ///
4384 /// This intrinsic has no corresponding instruction.
4385 ///
4386 /// \param __a
4387 ///    A 256-bit floating-point vector of [8 x float].
4388 /// \returns A 256-bit floating-point vector of [4 x double] containing the same
4389 ///    bitwise pattern as the parameter.
4390 static __inline __m256d __DEFAULT_FN_ATTRS
4391 _mm256_castps_pd(__m256 __a)
4392 {
4393   return (__m256d)__a;
4394 }
4395 
4396 /// Casts a 256-bit floating-point vector of [8 x float] into a 256-bit
4397 ///    integer vector.
4398 ///
4399 /// \headerfile <x86intrin.h>
4400 ///
4401 /// This intrinsic has no corresponding instruction.
4402 ///
4403 /// \param __a
4404 ///    A 256-bit floating-point vector of [8 x float].
4405 /// \returns A 256-bit integer vector containing the same bitwise pattern as the
4406 ///    parameter.
4407 static __inline __m256i __DEFAULT_FN_ATTRS
4408 _mm256_castps_si256(__m256 __a)
4409 {
4410   return (__m256i)__a;
4411 }
4412 
4413 /// Casts a 256-bit integer vector into a 256-bit floating-point vector
4414 ///    of [8 x float].
4415 ///
4416 /// \headerfile <x86intrin.h>
4417 ///
4418 /// This intrinsic has no corresponding instruction.
4419 ///
4420 /// \param __a
4421 ///    A 256-bit integer vector.
4422 /// \returns A 256-bit floating-point vector of [8 x float] containing the same
4423 ///    bitwise pattern as the parameter.
4424 static __inline __m256 __DEFAULT_FN_ATTRS
4425 _mm256_castsi256_ps(__m256i __a)
4426 {
4427   return (__m256)__a;
4428 }
4429 
4430 /// Casts a 256-bit integer vector into a 256-bit floating-point vector
4431 ///    of [4 x double].
4432 ///
4433 /// \headerfile <x86intrin.h>
4434 ///
4435 /// This intrinsic has no corresponding instruction.
4436 ///
4437 /// \param __a
4438 ///    A 256-bit integer vector.
4439 /// \returns A 256-bit floating-point vector of [4 x double] containing the same
4440 ///    bitwise pattern as the parameter.
4441 static __inline __m256d __DEFAULT_FN_ATTRS
4442 _mm256_castsi256_pd(__m256i __a)
4443 {
4444   return (__m256d)__a;
4445 }
4446 
4447 /// Returns the lower 128 bits of a 256-bit floating-point vector of
4448 ///    [4 x double] as a 128-bit floating-point vector of [2 x double].
4449 ///
4450 /// \headerfile <x86intrin.h>
4451 ///
4452 /// This intrinsic has no corresponding instruction.
4453 ///
4454 /// \param __a
4455 ///    A 256-bit floating-point vector of [4 x double].
4456 /// \returns A 128-bit floating-point vector of [2 x double] containing the
4457 ///    lower 128 bits of the parameter.
4458 static __inline __m128d __DEFAULT_FN_ATTRS
4459 _mm256_castpd256_pd128(__m256d __a)
4460 {
4461   return __builtin_shufflevector((__v4df)__a, (__v4df)__a, 0, 1);
4462 }
4463 
4464 /// Returns the lower 128 bits of a 256-bit floating-point vector of
4465 ///    [8 x float] as a 128-bit floating-point vector of [4 x float].
4466 ///
4467 /// \headerfile <x86intrin.h>
4468 ///
4469 /// This intrinsic has no corresponding instruction.
4470 ///
4471 /// \param __a
4472 ///    A 256-bit floating-point vector of [8 x float].
4473 /// \returns A 128-bit floating-point vector of [4 x float] containing the
4474 ///    lower 128 bits of the parameter.
4475 static __inline __m128 __DEFAULT_FN_ATTRS
4476 _mm256_castps256_ps128(__m256 __a)
4477 {
4478   return __builtin_shufflevector((__v8sf)__a, (__v8sf)__a, 0, 1, 2, 3);
4479 }
4480 
4481 /// Truncates a 256-bit integer vector into a 128-bit integer vector.
4482 ///
4483 /// \headerfile <x86intrin.h>
4484 ///
4485 /// This intrinsic has no corresponding instruction.
4486 ///
4487 /// \param __a
4488 ///    A 256-bit integer vector.
4489 /// \returns A 128-bit integer vector containing the lower 128 bits of the
4490 ///    parameter.
4491 static __inline __m128i __DEFAULT_FN_ATTRS
4492 _mm256_castsi256_si128(__m256i __a)
4493 {
4494   return __builtin_shufflevector((__v4di)__a, (__v4di)__a, 0, 1);
4495 }
4496 
4497 /// Constructs a 256-bit floating-point vector of [4 x double] from a
4498 ///    128-bit floating-point vector of [2 x double].
4499 ///
4500 ///    The lower 128 bits contain the value of the source vector. The contents
4501 ///    of the upper 128 bits are undefined.
4502 ///
4503 /// \headerfile <x86intrin.h>
4504 ///
4505 /// This intrinsic has no corresponding instruction.
4506 ///
4507 /// \param __a
4508 ///    A 128-bit vector of [2 x double].
4509 /// \returns A 256-bit floating-point vector of [4 x double]. The lower 128 bits
4510 ///    contain the value of the parameter. The contents of the upper 128 bits
4511 ///    are undefined.
4512 static __inline __m256d __DEFAULT_FN_ATTRS
4513 _mm256_castpd128_pd256(__m128d __a)
4514 {
4515   return __builtin_shufflevector(
4516       (__v2df)__a, (__v2df)__builtin_nondeterministic_value(__a), 0, 1, 2, 3);
4517 }
4518 
4519 /// Constructs a 256-bit floating-point vector of [8 x float] from a
4520 ///    128-bit floating-point vector of [4 x float].
4521 ///
4522 ///    The lower 128 bits contain the value of the source vector. The contents
4523 ///    of the upper 128 bits are undefined.
4524 ///
4525 /// \headerfile <x86intrin.h>
4526 ///
4527 /// This intrinsic has no corresponding instruction.
4528 ///
4529 /// \param __a
4530 ///    A 128-bit vector of [4 x float].
4531 /// \returns A 256-bit floating-point vector of [8 x float]. The lower 128 bits
4532 ///    contain the value of the parameter. The contents of the upper 128 bits
4533 ///    are undefined.
4534 static __inline __m256 __DEFAULT_FN_ATTRS
4535 _mm256_castps128_ps256(__m128 __a)
4536 {
4537   return __builtin_shufflevector((__v4sf)__a,
4538                                  (__v4sf)__builtin_nondeterministic_value(__a),
4539                                  0, 1, 2, 3, 4, 5, 6, 7);
4540 }
4541 
4542 /// Constructs a 256-bit integer vector from a 128-bit integer vector.
4543 ///
4544 ///    The lower 128 bits contain the value of the source vector. The contents
4545 ///    of the upper 128 bits are undefined.
4546 ///
4547 /// \headerfile <x86intrin.h>
4548 ///
4549 /// This intrinsic has no corresponding instruction.
4550 ///
4551 /// \param __a
4552 ///    A 128-bit integer vector.
4553 /// \returns A 256-bit integer vector. The lower 128 bits contain the value of
4554 ///    the parameter. The contents of the upper 128 bits are undefined.
4555 static __inline __m256i __DEFAULT_FN_ATTRS
4556 _mm256_castsi128_si256(__m128i __a)
4557 {
4558   return __builtin_shufflevector(
4559       (__v2di)__a, (__v2di)__builtin_nondeterministic_value(__a), 0, 1, 2, 3);
4560 }
4561 
4562 /// Constructs a 256-bit floating-point vector of [4 x double] from a
4563 ///    128-bit floating-point vector of [2 x double]. The lower 128 bits
4564 ///    contain the value of the source vector. The upper 128 bits are set
4565 ///    to zero.
4566 ///
4567 /// \headerfile <x86intrin.h>
4568 ///
4569 /// This intrinsic has no corresponding instruction.
4570 ///
4571 /// \param __a
4572 ///    A 128-bit vector of [2 x double].
4573 /// \returns A 256-bit floating-point vector of [4 x double]. The lower 128 bits
4574 ///    contain the value of the parameter. The upper 128 bits are set to zero.
4575 static __inline __m256d __DEFAULT_FN_ATTRS
4576 _mm256_zextpd128_pd256(__m128d __a)
4577 {
4578   return __builtin_shufflevector((__v2df)__a, (__v2df)_mm_setzero_pd(), 0, 1, 2, 3);
4579 }
4580 
4581 /// Constructs a 256-bit floating-point vector of [8 x float] from a
4582 ///    128-bit floating-point vector of [4 x float]. The lower 128 bits contain
4583 ///    the value of the source vector. The upper 128 bits are set to zero.
4584 ///
4585 /// \headerfile <x86intrin.h>
4586 ///
4587 /// This intrinsic has no corresponding instruction.
4588 ///
4589 /// \param __a
4590 ///    A 128-bit vector of [4 x float].
4591 /// \returns A 256-bit floating-point vector of [8 x float]. The lower 128 bits
4592 ///    contain the value of the parameter. The upper 128 bits are set to zero.
4593 static __inline __m256 __DEFAULT_FN_ATTRS
4594 _mm256_zextps128_ps256(__m128 __a)
4595 {
4596   return __builtin_shufflevector((__v4sf)__a, (__v4sf)_mm_setzero_ps(), 0, 1, 2, 3, 4, 5, 6, 7);
4597 }
4598 
4599 /// Constructs a 256-bit integer vector from a 128-bit integer vector.
4600 ///    The lower 128 bits contain the value of the source vector. The upper
4601 ///    128 bits are set to zero.
4602 ///
4603 /// \headerfile <x86intrin.h>
4604 ///
4605 /// This intrinsic has no corresponding instruction.
4606 ///
4607 /// \param __a
4608 ///    A 128-bit integer vector.
4609 /// \returns A 256-bit integer vector. The lower 128 bits contain the value of
4610 ///    the parameter. The upper 128 bits are set to zero.
4611 static __inline __m256i __DEFAULT_FN_ATTRS
4612 _mm256_zextsi128_si256(__m128i __a)
4613 {
4614   return __builtin_shufflevector((__v2di)__a, (__v2di)_mm_setzero_si128(), 0, 1, 2, 3);
4615 }
4616 
4617 /*
4618    Vector insert.
4619    We use macros rather than inlines because we only want to accept
4620    invocations where the immediate M is a constant expression.
4621 */
4622 /// Constructs a new 256-bit vector of [8 x float] by first duplicating
4623 ///    a 256-bit vector of [8 x float] given in the first parameter, and then
4624 ///    replacing either the upper or the lower 128 bits with the contents of a
4625 ///    128-bit vector of [4 x float] in the second parameter.
4626 ///
4627 ///    The immediate integer parameter determines between the upper or the lower
4628 ///    128 bits.
4629 ///
4630 /// \headerfile <x86intrin.h>
4631 ///
4632 /// \code
4633 /// __m256 _mm256_insertf128_ps(__m256 V1, __m128 V2, const int M);
4634 /// \endcode
4635 ///
4636 /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction.
4637 ///
4638 /// \param V1
4639 ///    A 256-bit vector of [8 x float]. This vector is copied to the result
4640 ///    first, and then either the upper or the lower 128 bits of the result will
4641 ///    be replaced by the contents of \a V2.
4642 /// \param V2
4643 ///    A 128-bit vector of [4 x float]. The contents of this parameter are
4644 ///    written to either the upper or the lower 128 bits of the result depending
4645 ///    on the value of parameter \a M.
4646 /// \param M
4647 ///    An immediate integer. The least significant bit determines how the values
4648 ///    from the two parameters are interleaved: \n
4649 ///    If bit [0] of \a M is 0, \a V2 are copied to bits [127:0] of the result,
4650 ///    and bits [255:128] of \a V1 are copied to bits [255:128] of the
4651 ///    result. \n
4652 ///    If bit [0] of \a M is 1, \a V2 are copied to bits [255:128] of the
4653 ///    result, and bits [127:0] of \a V1 are copied to bits [127:0] of the
4654 ///    result.
4655 /// \returns A 256-bit vector of [8 x float] containing the interleaved values.
4656 #define _mm256_insertf128_ps(V1, V2, M) \
4657   ((__m256)__builtin_ia32_vinsertf128_ps256((__v8sf)(__m256)(V1), \
4658                                             (__v4sf)(__m128)(V2), (int)(M)))
4659 
4660 /// Constructs a new 256-bit vector of [4 x double] by first duplicating
4661 ///    a 256-bit vector of [4 x double] given in the first parameter, and then
4662 ///    replacing either the upper or the lower 128 bits with the contents of a
4663 ///    128-bit vector of [2 x double] in the second parameter.
4664 ///
4665 ///    The immediate integer parameter determines between the upper or the lower
4666 ///    128 bits.
4667 ///
4668 /// \headerfile <x86intrin.h>
4669 ///
4670 /// \code
4671 /// __m256d _mm256_insertf128_pd(__m256d V1, __m128d V2, const int M);
4672 /// \endcode
4673 ///
4674 /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction.
4675 ///
4676 /// \param V1
4677 ///    A 256-bit vector of [4 x double]. This vector is copied to the result
4678 ///    first, and then either the upper or the lower 128 bits of the result will
4679 ///    be replaced by the contents of \a V2.
4680 /// \param V2
4681 ///    A 128-bit vector of [2 x double]. The contents of this parameter are
4682 ///    written to either the upper or the lower 128 bits of the result depending
4683 ///    on the value of parameter \a M.
4684 /// \param M
4685 ///    An immediate integer. The least significant bit determines how the values
4686 ///    from the two parameters are interleaved: \n
4687 ///    If bit [0] of \a M is 0, \a V2 are copied to bits [127:0] of the result,
4688 ///    and bits [255:128] of \a V1 are copied to bits [255:128] of the
4689 ///    result. \n
4690 ///    If bit [0] of \a M is 1, \a V2 are copied to bits [255:128] of the
4691 ///    result, and bits [127:0] of \a V1 are copied to bits [127:0] of the
4692 ///    result.
4693 /// \returns A 256-bit vector of [4 x double] containing the interleaved values.
4694 #define _mm256_insertf128_pd(V1, V2, M) \
4695   ((__m256d)__builtin_ia32_vinsertf128_pd256((__v4df)(__m256d)(V1), \
4696                                              (__v2df)(__m128d)(V2), (int)(M)))
4697 
4698 /// Constructs a new 256-bit integer vector by first duplicating a
4699 ///    256-bit integer vector given in the first parameter, and then replacing
4700 ///    either the upper or the lower 128 bits with the contents of a 128-bit
4701 ///    integer vector in the second parameter.
4702 ///
4703 ///    The immediate integer parameter determines between the upper or the lower
4704 ///    128 bits.
4705 ///
4706 /// \headerfile <x86intrin.h>
4707 ///
4708 /// \code
4709 /// __m256i _mm256_insertf128_si256(__m256i V1, __m128i V2, const int M);
4710 /// \endcode
4711 ///
4712 /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction.
4713 ///
4714 /// \param V1
4715 ///    A 256-bit integer vector. This vector is copied to the result first, and
4716 ///    then either the upper or the lower 128 bits of the result will be
4717 ///    replaced by the contents of \a V2.
4718 /// \param V2
4719 ///    A 128-bit integer vector. The contents of this parameter are written to
4720 ///    either the upper or the lower 128 bits of the result depending on the
4721 ///     value of parameter \a M.
4722 /// \param M
4723 ///    An immediate integer. The least significant bit determines how the values
4724 ///    from the two parameters are interleaved: \n
4725 ///    If bit [0] of \a M is 0, \a V2 are copied to bits [127:0] of the result,
4726 ///    and bits [255:128] of \a V1 are copied to bits [255:128] of the
4727 ///    result. \n
4728 ///    If bit [0] of \a M is 1, \a V2 are copied to bits [255:128] of the
4729 ///    result, and bits [127:0] of \a V1 are copied to bits [127:0] of the
4730 ///    result.
4731 /// \returns A 256-bit integer vector containing the interleaved values.
4732 #define _mm256_insertf128_si256(V1, V2, M) \
4733   ((__m256i)__builtin_ia32_vinsertf128_si256((__v8si)(__m256i)(V1), \
4734                                              (__v4si)(__m128i)(V2), (int)(M)))
4735 
4736 /*
4737    Vector extract.
4738    We use macros rather than inlines because we only want to accept
4739    invocations where the immediate M is a constant expression.
4740 */
4741 /// Extracts either the upper or the lower 128 bits from a 256-bit vector
4742 ///    of [8 x float], as determined by the immediate integer parameter, and
4743 ///    returns the extracted bits as a 128-bit vector of [4 x float].
4744 ///
4745 /// \headerfile <x86intrin.h>
4746 ///
4747 /// \code
4748 /// __m128 _mm256_extractf128_ps(__m256 V, const int M);
4749 /// \endcode
4750 ///
4751 /// This intrinsic corresponds to the <c> VEXTRACTF128 </c> instruction.
4752 ///
4753 /// \param V
4754 ///    A 256-bit vector of [8 x float].
4755 /// \param M
4756 ///    An immediate integer. The least significant bit determines which bits are
4757 ///    extracted from the first parameter: \n
4758 ///    If bit [0] of \a M is 0, bits [127:0] of \a V are copied to the
4759 ///    result. \n
4760 ///    If bit [0] of \a M is 1, bits [255:128] of \a V are copied to the result.
4761 /// \returns A 128-bit vector of [4 x float] containing the extracted bits.
4762 #define _mm256_extractf128_ps(V, M) \
4763   ((__m128)__builtin_ia32_vextractf128_ps256((__v8sf)(__m256)(V), (int)(M)))
4764 
4765 /// Extracts either the upper or the lower 128 bits from a 256-bit vector
4766 ///    of [4 x double], as determined by the immediate integer parameter, and
4767 ///    returns the extracted bits as a 128-bit vector of [2 x double].
4768 ///
4769 /// \headerfile <x86intrin.h>
4770 ///
4771 /// \code
4772 /// __m128d _mm256_extractf128_pd(__m256d V, const int M);
4773 /// \endcode
4774 ///
4775 /// This intrinsic corresponds to the <c> VEXTRACTF128 </c> instruction.
4776 ///
4777 /// \param V
4778 ///    A 256-bit vector of [4 x double].
4779 /// \param M
4780 ///    An immediate integer. The least significant bit determines which bits are
4781 ///    extracted from the first parameter: \n
4782 ///    If bit [0] of \a M is 0, bits [127:0] of \a V are copied to the
4783 ///    result. \n
4784 ///    If bit [0] of \a M is 1, bits [255:128] of \a V are copied to the result.
4785 /// \returns A 128-bit vector of [2 x double] containing the extracted bits.
4786 #define _mm256_extractf128_pd(V, M) \
4787   ((__m128d)__builtin_ia32_vextractf128_pd256((__v4df)(__m256d)(V), (int)(M)))
4788 
4789 /// Extracts either the upper or the lower 128 bits from a 256-bit
4790 ///    integer vector, as determined by the immediate integer parameter, and
4791 ///    returns the extracted bits as a 128-bit integer vector.
4792 ///
4793 /// \headerfile <x86intrin.h>
4794 ///
4795 /// \code
4796 /// __m128i _mm256_extractf128_si256(__m256i V, const int M);
4797 /// \endcode
4798 ///
4799 /// This intrinsic corresponds to the <c> VEXTRACTF128 </c> instruction.
4800 ///
4801 /// \param V
4802 ///    A 256-bit integer vector.
4803 /// \param M
4804 ///    An immediate integer. The least significant bit determines which bits are
4805 ///    extracted from the first parameter:  \n
4806 ///    If bit [0] of \a M is 0, bits [127:0] of \a V are copied to the
4807 ///    result. \n
4808 ///    If bit [0] of \a M is 1, bits [255:128] of \a V are copied to the result.
4809 /// \returns A 128-bit integer vector containing the extracted bits.
4810 #define _mm256_extractf128_si256(V, M) \
4811   ((__m128i)__builtin_ia32_vextractf128_si256((__v8si)(__m256i)(V), (int)(M)))
4812 
4813 /// Constructs a 256-bit floating-point vector of [8 x float] by
4814 ///    concatenating two 128-bit floating-point vectors of [4 x float].
4815 ///
4816 /// \headerfile <x86intrin.h>
4817 ///
4818 /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction.
4819 ///
4820 /// \param __hi
4821 ///    A 128-bit floating-point vector of [4 x float] to be copied to the upper
4822 ///    128 bits of the result.
4823 /// \param __lo
4824 ///    A 128-bit floating-point vector of [4 x float] to be copied to the lower
4825 ///    128 bits of the result.
4826 /// \returns A 256-bit floating-point vector of [8 x float] containing the
4827 ///    concatenated result.
4828 static __inline __m256 __DEFAULT_FN_ATTRS
4829 _mm256_set_m128 (__m128 __hi, __m128 __lo)
4830 {
4831   return (__m256) __builtin_shufflevector((__v4sf)__lo, (__v4sf)__hi, 0, 1, 2, 3, 4, 5, 6, 7);
4832 }
4833 
4834 /// Constructs a 256-bit floating-point vector of [4 x double] by
4835 ///    concatenating two 128-bit floating-point vectors of [2 x double].
4836 ///
4837 /// \headerfile <x86intrin.h>
4838 ///
4839 /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction.
4840 ///
4841 /// \param __hi
4842 ///    A 128-bit floating-point vector of [2 x double] to be copied to the upper
4843 ///    128 bits of the result.
4844 /// \param __lo
4845 ///    A 128-bit floating-point vector of [2 x double] to be copied to the lower
4846 ///    128 bits of the result.
4847 /// \returns A 256-bit floating-point vector of [4 x double] containing the
4848 ///    concatenated result.
4849 static __inline __m256d __DEFAULT_FN_ATTRS
4850 _mm256_set_m128d (__m128d __hi, __m128d __lo)
4851 {
4852   return (__m256d) __builtin_shufflevector((__v2df)__lo, (__v2df)__hi, 0, 1, 2, 3);
4853 }
4854 
4855 /// Constructs a 256-bit integer vector by concatenating two 128-bit
4856 ///    integer vectors.
4857 ///
4858 /// \headerfile <x86intrin.h>
4859 ///
4860 /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction.
4861 ///
4862 /// \param __hi
4863 ///    A 128-bit integer vector to be copied to the upper 128 bits of the
4864 ///    result.
4865 /// \param __lo
4866 ///    A 128-bit integer vector to be copied to the lower 128 bits of the
4867 ///    result.
4868 /// \returns A 256-bit integer vector containing the concatenated result.
4869 static __inline __m256i __DEFAULT_FN_ATTRS
4870 _mm256_set_m128i (__m128i __hi, __m128i __lo)
4871 {
4872   return (__m256i) __builtin_shufflevector((__v2di)__lo, (__v2di)__hi, 0, 1, 2, 3);
4873 }
4874 
4875 /// Constructs a 256-bit floating-point vector of [8 x float] by
4876 ///    concatenating two 128-bit floating-point vectors of [4 x float]. This is
4877 ///    similar to _mm256_set_m128, but the order of the input parameters is
4878 ///    swapped.
4879 ///
4880 /// \headerfile <x86intrin.h>
4881 ///
4882 /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction.
4883 ///
4884 /// \param __lo
4885 ///    A 128-bit floating-point vector of [4 x float] to be copied to the lower
4886 ///    128 bits of the result.
4887 /// \param __hi
4888 ///    A 128-bit floating-point vector of [4 x float] to be copied to the upper
4889 ///    128 bits of the result.
4890 /// \returns A 256-bit floating-point vector of [8 x float] containing the
4891 ///    concatenated result.
4892 static __inline __m256 __DEFAULT_FN_ATTRS
4893 _mm256_setr_m128 (__m128 __lo, __m128 __hi)
4894 {
4895   return _mm256_set_m128(__hi, __lo);
4896 }
4897 
4898 /// Constructs a 256-bit floating-point vector of [4 x double] by
4899 ///    concatenating two 128-bit floating-point vectors of [2 x double]. This is
4900 ///    similar to _mm256_set_m128d, but the order of the input parameters is
4901 ///    swapped.
4902 ///
4903 /// \headerfile <x86intrin.h>
4904 ///
4905 /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction.
4906 ///
4907 /// \param __lo
4908 ///    A 128-bit floating-point vector of [2 x double] to be copied to the lower
4909 ///    128 bits of the result.
4910 /// \param __hi
4911 ///    A 128-bit floating-point vector of [2 x double] to be copied to the upper
4912 ///    128 bits of the result.
4913 /// \returns A 256-bit floating-point vector of [4 x double] containing the
4914 ///    concatenated result.
4915 static __inline __m256d __DEFAULT_FN_ATTRS
4916 _mm256_setr_m128d (__m128d __lo, __m128d __hi)
4917 {
4918   return (__m256d)_mm256_set_m128d(__hi, __lo);
4919 }
4920 
4921 /// Constructs a 256-bit integer vector by concatenating two 128-bit
4922 ///    integer vectors. This is similar to _mm256_set_m128i, but the order of
4923 ///    the input parameters is swapped.
4924 ///
4925 /// \headerfile <x86intrin.h>
4926 ///
4927 /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction.
4928 ///
4929 /// \param __lo
4930 ///    A 128-bit integer vector to be copied to the lower 128 bits of the
4931 ///    result.
4932 /// \param __hi
4933 ///    A 128-bit integer vector to be copied to the upper 128 bits of the
4934 ///    result.
4935 /// \returns A 256-bit integer vector containing the concatenated result.
4936 static __inline __m256i __DEFAULT_FN_ATTRS
4937 _mm256_setr_m128i (__m128i __lo, __m128i __hi)
4938 {
4939   return (__m256i)_mm256_set_m128i(__hi, __lo);
4940 }
4941 
4942 /* SIMD load ops (unaligned) */
4943 /// Loads two 128-bit floating-point vectors of [4 x float] from
4944 ///    unaligned memory locations and constructs a 256-bit floating-point vector
4945 ///    of [8 x float] by concatenating the two 128-bit vectors.
4946 ///
4947 /// \headerfile <x86intrin.h>
4948 ///
4949 /// This intrinsic corresponds to load instructions followed by the
4950 ///   <c> VINSERTF128 </c> instruction.
4951 ///
4952 /// \param __addr_hi
4953 ///    A pointer to a 128-bit memory location containing 4 consecutive
4954 ///    single-precision floating-point values. These values are to be copied to
4955 ///    bits[255:128] of the result. The address of the memory location does not
4956 ///    have to be aligned.
4957 /// \param __addr_lo
4958 ///    A pointer to a 128-bit memory location containing 4 consecutive
4959 ///    single-precision floating-point values. These values are to be copied to
4960 ///    bits[127:0] of the result. The address of the memory location does not
4961 ///    have to be aligned.
4962 /// \returns A 256-bit floating-point vector of [8 x float] containing the
4963 ///    concatenated result.
4964 static __inline __m256 __DEFAULT_FN_ATTRS
4965 _mm256_loadu2_m128(float const *__addr_hi, float const *__addr_lo)
4966 {
4967   return _mm256_set_m128(_mm_loadu_ps(__addr_hi), _mm_loadu_ps(__addr_lo));
4968 }
4969 
4970 /// Loads two 128-bit floating-point vectors of [2 x double] from
4971 ///    unaligned memory locations and constructs a 256-bit floating-point vector
4972 ///    of [4 x double] by concatenating the two 128-bit vectors.
4973 ///
4974 /// \headerfile <x86intrin.h>
4975 ///
4976 /// This intrinsic corresponds to load instructions followed by the
4977 ///   <c> VINSERTF128 </c> instruction.
4978 ///
4979 /// \param __addr_hi
4980 ///    A pointer to a 128-bit memory location containing two consecutive
4981 ///    double-precision floating-point values. These values are to be copied to
4982 ///    bits[255:128] of the result. The address of the memory location does not
4983 ///    have to be aligned.
4984 /// \param __addr_lo
4985 ///    A pointer to a 128-bit memory location containing two consecutive
4986 ///    double-precision floating-point values. These values are to be copied to
4987 ///    bits[127:0] of the result. The address of the memory location does not
4988 ///    have to be aligned.
4989 /// \returns A 256-bit floating-point vector of [4 x double] containing the
4990 ///    concatenated result.
4991 static __inline __m256d __DEFAULT_FN_ATTRS
4992 _mm256_loadu2_m128d(double const *__addr_hi, double const *__addr_lo)
4993 {
4994   return _mm256_set_m128d(_mm_loadu_pd(__addr_hi), _mm_loadu_pd(__addr_lo));
4995 }
4996 
4997 /// Loads two 128-bit integer vectors from unaligned memory locations and
4998 ///    constructs a 256-bit integer vector by concatenating the two 128-bit
4999 ///    vectors.
5000 ///
5001 /// \headerfile <x86intrin.h>
5002 ///
5003 /// This intrinsic corresponds to load instructions followed by the
5004 ///   <c> VINSERTF128 </c> instruction.
5005 ///
5006 /// \param __addr_hi
5007 ///    A pointer to a 128-bit memory location containing a 128-bit integer
5008 ///    vector. This vector is to be copied to bits[255:128] of the result. The
5009 ///    address of the memory location does not have to be aligned.
5010 /// \param __addr_lo
5011 ///    A pointer to a 128-bit memory location containing a 128-bit integer
5012 ///    vector. This vector is to be copied to bits[127:0] of the result. The
5013 ///    address of the memory location does not have to be aligned.
5014 /// \returns A 256-bit integer vector containing the concatenated result.
5015 static __inline __m256i __DEFAULT_FN_ATTRS
5016 _mm256_loadu2_m128i(__m128i_u const *__addr_hi, __m128i_u const *__addr_lo)
5017 {
5018    return _mm256_set_m128i(_mm_loadu_si128(__addr_hi), _mm_loadu_si128(__addr_lo));
5019 }
5020 
5021 /* SIMD store ops (unaligned) */
5022 /// Stores the upper and lower 128 bits of a 256-bit floating-point
5023 ///    vector of [8 x float] into two different unaligned memory locations.
5024 ///
5025 /// \headerfile <x86intrin.h>
5026 ///
5027 /// This intrinsic corresponds to the <c> VEXTRACTF128 </c> instruction and the
5028 ///   store instructions.
5029 ///
5030 /// \param __addr_hi
5031 ///    A pointer to a 128-bit memory location. Bits[255:128] of \a __a are to be
5032 ///    copied to this memory location. The address of this memory location does
5033 ///    not have to be aligned.
5034 /// \param __addr_lo
5035 ///    A pointer to a 128-bit memory location. Bits[127:0] of \a __a are to be
5036 ///    copied to this memory location. The address of this memory location does
5037 ///    not have to be aligned.
5038 /// \param __a
5039 ///    A 256-bit floating-point vector of [8 x float].
5040 static __inline void __DEFAULT_FN_ATTRS
5041 _mm256_storeu2_m128(float *__addr_hi, float *__addr_lo, __m256 __a)
5042 {
5043   __m128 __v128;
5044 
5045   __v128 = _mm256_castps256_ps128(__a);
5046   _mm_storeu_ps(__addr_lo, __v128);
5047   __v128 = _mm256_extractf128_ps(__a, 1);
5048   _mm_storeu_ps(__addr_hi, __v128);
5049 }
5050 
5051 /// Stores the upper and lower 128 bits of a 256-bit floating-point
5052 ///    vector of [4 x double] into two different unaligned memory locations.
5053 ///
5054 /// \headerfile <x86intrin.h>
5055 ///
5056 /// This intrinsic corresponds to the <c> VEXTRACTF128 </c> instruction and the
5057 ///   store instructions.
5058 ///
5059 /// \param __addr_hi
5060 ///    A pointer to a 128-bit memory location. Bits[255:128] of \a __a are to be
5061 ///    copied to this memory location. The address of this memory location does
5062 ///    not have to be aligned.
5063 /// \param __addr_lo
5064 ///    A pointer to a 128-bit memory location. Bits[127:0] of \a __a are to be
5065 ///    copied to this memory location. The address of this memory location does
5066 ///    not have to be aligned.
5067 /// \param __a
5068 ///    A 256-bit floating-point vector of [4 x double].
5069 static __inline void __DEFAULT_FN_ATTRS
5070 _mm256_storeu2_m128d(double *__addr_hi, double *__addr_lo, __m256d __a)
5071 {
5072   __m128d __v128;
5073 
5074   __v128 = _mm256_castpd256_pd128(__a);
5075   _mm_storeu_pd(__addr_lo, __v128);
5076   __v128 = _mm256_extractf128_pd(__a, 1);
5077   _mm_storeu_pd(__addr_hi, __v128);
5078 }
5079 
5080 /// Stores the upper and lower 128 bits of a 256-bit integer vector into
5081 ///    two different unaligned memory locations.
5082 ///
5083 /// \headerfile <x86intrin.h>
5084 ///
5085 /// This intrinsic corresponds to the <c> VEXTRACTF128 </c> instruction and the
5086 ///   store instructions.
5087 ///
5088 /// \param __addr_hi
5089 ///    A pointer to a 128-bit memory location. Bits[255:128] of \a __a are to be
5090 ///    copied to this memory location. The address of this memory location does
5091 ///    not have to be aligned.
5092 /// \param __addr_lo
5093 ///    A pointer to a 128-bit memory location. Bits[127:0] of \a __a are to be
5094 ///    copied to this memory location. The address of this memory location does
5095 ///    not have to be aligned.
5096 /// \param __a
5097 ///    A 256-bit integer vector.
5098 static __inline void __DEFAULT_FN_ATTRS
5099 _mm256_storeu2_m128i(__m128i_u *__addr_hi, __m128i_u *__addr_lo, __m256i __a)
5100 {
5101   __m128i __v128;
5102 
5103   __v128 = _mm256_castsi256_si128(__a);
5104   _mm_storeu_si128(__addr_lo, __v128);
5105   __v128 = _mm256_extractf128_si256(__a, 1);
5106   _mm_storeu_si128(__addr_hi, __v128);
5107 }
5108 
5109 #undef __DEFAULT_FN_ATTRS
5110 #undef __DEFAULT_FN_ATTRS128
5111 
5112 #endif /* __AVXINTRIN_H */
5113