1 | /*===---- avxintrin.h - AVX intrinsics -------------------------------------=== |
---|---|
2 | * |
3 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
4 | * of this software and associated documentation files (the "Software"), to deal |
5 | * in the Software without restriction, including without limitation the rights |
6 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
7 | * copies of the Software, and to permit persons to whom the Software is |
8 | * furnished to do so, subject to the following conditions: |
9 | * |
10 | * The above copyright notice and this permission notice shall be included in |
11 | * all copies or substantial portions of the Software. |
12 | * |
13 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
14 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
15 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
16 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
17 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
18 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
19 | * THE SOFTWARE. |
20 | * |
21 | *===-----------------------------------------------------------------------=== |
22 | */ |
23 | |
24 | #ifndef __IMMINTRIN_H |
25 | #error "Never use <avxintrin.h> directly; include <immintrin.h> instead." |
26 | #endif |
27 | |
28 | #ifndef __AVXINTRIN_H |
29 | #define __AVXINTRIN_H |
30 | |
31 | typedef double __v4df __attribute__ ((__vector_size__ (32))); |
32 | typedef float __v8sf __attribute__ ((__vector_size__ (32))); |
33 | typedef long long __v4di __attribute__ ((__vector_size__ (32))); |
34 | typedef int __v8si __attribute__ ((__vector_size__ (32))); |
35 | typedef short __v16hi __attribute__ ((__vector_size__ (32))); |
36 | typedef char __v32qi __attribute__ ((__vector_size__ (32))); |
37 | |
38 | /* Unsigned types */ |
39 | typedef unsigned long long __v4du __attribute__ ((__vector_size__ (32))); |
40 | typedef unsigned int __v8su __attribute__ ((__vector_size__ (32))); |
41 | typedef unsigned short __v16hu __attribute__ ((__vector_size__ (32))); |
42 | typedef unsigned char __v32qu __attribute__ ((__vector_size__ (32))); |
43 | |
44 | /* We need an explicitly signed variant for char. Note that this shouldn't |
45 | * appear in the interface though. */ |
46 | typedef signed char __v32qs __attribute__((__vector_size__(32))); |
47 | |
48 | typedef float __m256 __attribute__ ((__vector_size__ (32), __aligned__(32))); |
49 | typedef double __m256d __attribute__((__vector_size__(32), __aligned__(32))); |
50 | typedef long long __m256i __attribute__((__vector_size__(32), __aligned__(32))); |
51 | |
52 | typedef float __m256_u __attribute__ ((__vector_size__ (32), __aligned__(1))); |
53 | typedef double __m256d_u __attribute__((__vector_size__(32), __aligned__(1))); |
54 | typedef long long __m256i_u __attribute__((__vector_size__(32), __aligned__(1))); |
55 | |
56 | /* Define the default attributes for the functions in this file. */ |
57 | #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx"), __min_vector_width__(256))) |
58 | #define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("avx"), __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. |
1522 | /// \returns A 256-bit vector of [8 x float] containing the shuffled values. |
1523 | #define _mm256_shuffle_ps(a, b, mask) \ |
1524 | (__m256)__builtin_ia32_shufps256((__v8sf)(__m256)(a), \ |
1525 | (__v8sf)(__m256)(b), (int)(mask)) |
1526 | |
1527 | /// Selects four double-precision values from the 256-bit operands of |
1528 | /// [4 x double], as specified by the immediate value operand. |
1529 | /// |
1530 | /// The selected elements from the first 256-bit operand are copied to bits |
1531 | /// [63:0] and bits [191:128] in the destination, and the selected elements |
1532 | /// from the second 256-bit operand are copied to bits [127:64] and bits |
1533 | /// [255:192] in the destination. For example, if bits [3:0] of the immediate |
1534 | /// operand contain a value of 0xF, the 256-bit destination vector would |
1535 | /// contain the following values: b[3], a[3], b[1], a[1]. |
1536 | /// |
1537 | /// \headerfile <x86intrin.h> |
1538 | /// |
1539 | /// \code |
1540 | /// __m256d _mm256_shuffle_pd(__m256d a, __m256d b, const int mask); |
1541 | /// \endcode |
1542 | /// |
1543 | /// This intrinsic corresponds to the <c> VSHUFPD </c> instruction. |
1544 | /// |
1545 | /// \param a |
1546 | /// A 256-bit vector of [4 x double]. |
1547 | /// \param b |
1548 | /// A 256-bit vector of [4 x double]. |
1549 | /// \param mask |
1550 | /// An immediate value containing 8-bit values specifying which elements to |
1551 | /// copy from \a a and \a b: \n |
1552 | /// Bit [0]=0: Bits [63:0] are copied from \a a to bits [63:0] of the |
1553 | /// destination. \n |
1554 | /// Bit [0]=1: Bits [127:64] are copied from \a a to bits [63:0] of the |
1555 | /// destination. \n |
1556 | /// Bit [1]=0: Bits [63:0] are copied from \a b to bits [127:64] of the |
1557 | /// destination. \n |
1558 | /// Bit [1]=1: Bits [127:64] are copied from \a b to bits [127:64] of the |
1559 | /// destination. \n |
1560 | /// Bit [2]=0: Bits [191:128] are copied from \a a to bits [191:128] of the |
1561 | /// destination. \n |
1562 | /// Bit [2]=1: Bits [255:192] are copied from \a a to bits [191:128] of the |
1563 | /// destination. \n |
1564 | /// Bit [3]=0: Bits [191:128] are copied from \a b to bits [255:192] of the |
1565 | /// destination. \n |
1566 | /// Bit [3]=1: Bits [255:192] are copied from \a b to bits [255:192] of the |
1567 | /// destination. |
1568 | /// \returns A 256-bit vector of [4 x double] containing the shuffled values. |
1569 | #define _mm256_shuffle_pd(a, b, mask) \ |
1570 | (__m256d)__builtin_ia32_shufpd256((__v4df)(__m256d)(a), \ |
1571 | (__v4df)(__m256d)(b), (int)(mask)) |
1572 | |
1573 | /* Compare */ |
1574 | #define _CMP_EQ_OQ 0x00 /* Equal (ordered, non-signaling) */ |
1575 | #define _CMP_LT_OS 0x01 /* Less-than (ordered, signaling) */ |
1576 | #define _CMP_LE_OS 0x02 /* Less-than-or-equal (ordered, signaling) */ |
1577 | #define _CMP_UNORD_Q 0x03 /* Unordered (non-signaling) */ |
1578 | #define _CMP_NEQ_UQ 0x04 /* Not-equal (unordered, non-signaling) */ |
1579 | #define _CMP_NLT_US 0x05 /* Not-less-than (unordered, signaling) */ |
1580 | #define _CMP_NLE_US 0x06 /* Not-less-than-or-equal (unordered, signaling) */ |
1581 | #define _CMP_ORD_Q 0x07 /* Ordered (non-signaling) */ |
1582 | #define _CMP_EQ_UQ 0x08 /* Equal (unordered, non-signaling) */ |
1583 | #define _CMP_NGE_US 0x09 /* Not-greater-than-or-equal (unordered, signaling) */ |
1584 | #define _CMP_NGT_US 0x0a /* Not-greater-than (unordered, signaling) */ |
1585 | #define _CMP_FALSE_OQ 0x0b /* False (ordered, non-signaling) */ |
1586 | #define _CMP_NEQ_OQ 0x0c /* Not-equal (ordered, non-signaling) */ |
1587 | #define _CMP_GE_OS 0x0d /* Greater-than-or-equal (ordered, signaling) */ |
1588 | #define _CMP_GT_OS 0x0e /* Greater-than (ordered, signaling) */ |
1589 | #define _CMP_TRUE_UQ 0x0f /* True (unordered, non-signaling) */ |
1590 | #define _CMP_EQ_OS 0x10 /* Equal (ordered, signaling) */ |
1591 | #define _CMP_LT_OQ 0x11 /* Less-than (ordered, non-signaling) */ |
1592 | #define _CMP_LE_OQ 0x12 /* Less-than-or-equal (ordered, non-signaling) */ |
1593 | #define _CMP_UNORD_S 0x13 /* Unordered (signaling) */ |
1594 | #define _CMP_NEQ_US 0x14 /* Not-equal (unordered, signaling) */ |
1595 | #define _CMP_NLT_UQ 0x15 /* Not-less-than (unordered, non-signaling) */ |
1596 | #define _CMP_NLE_UQ 0x16 /* Not-less-than-or-equal (unordered, non-signaling) */ |
1597 | #define _CMP_ORD_S 0x17 /* Ordered (signaling) */ |
1598 | #define _CMP_EQ_US 0x18 /* Equal (unordered, signaling) */ |
1599 | #define _CMP_NGE_UQ 0x19 /* Not-greater-than-or-equal (unordered, non-signaling) */ |
1600 | #define _CMP_NGT_UQ 0x1a /* Not-greater-than (unordered, non-signaling) */ |
1601 | #define _CMP_FALSE_OS 0x1b /* False (ordered, signaling) */ |
1602 | #define _CMP_NEQ_OS 0x1c /* Not-equal (ordered, signaling) */ |
1603 | #define _CMP_GE_OQ 0x1d /* Greater-than-or-equal (ordered, non-signaling) */ |
1604 | #define _CMP_GT_OQ 0x1e /* Greater-than (ordered, non-signaling) */ |
1605 | #define _CMP_TRUE_US 0x1f /* True (unordered, signaling) */ |
1606 | |
1607 | /// Compares each of the corresponding double-precision values of two |
1608 | /// 128-bit vectors of [2 x double], using the operation specified by the |
1609 | /// immediate integer operand. |
1610 | /// |
1611 | /// Returns a [2 x double] vector consisting of two doubles corresponding to |
1612 | /// the two comparison results: zero if the comparison is false, and all 1's |
1613 | /// if the comparison is true. |
1614 | /// |
1615 | /// \headerfile <x86intrin.h> |
1616 | /// |
1617 | /// \code |
1618 | /// __m128d _mm_cmp_pd(__m128d a, __m128d b, const int c); |
1619 | /// \endcode |
1620 | /// |
1621 | /// This intrinsic corresponds to the <c> VCMPPD </c> instruction. |
1622 | /// |
1623 | /// \param a |
1624 | /// A 128-bit vector of [2 x double]. |
1625 | /// \param b |
1626 | /// A 128-bit vector of [2 x double]. |
1627 | /// \param c |
1628 | /// An immediate integer operand, with bits [4:0] specifying which comparison |
1629 | /// operation to use: \n |
1630 | /// 0x00: Equal (ordered, non-signaling) \n |
1631 | /// 0x01: Less-than (ordered, signaling) \n |
1632 | /// 0x02: Less-than-or-equal (ordered, signaling) \n |
1633 | /// 0x03: Unordered (non-signaling) \n |
1634 | /// 0x04: Not-equal (unordered, non-signaling) \n |
1635 | /// 0x05: Not-less-than (unordered, signaling) \n |
1636 | /// 0x06: Not-less-than-or-equal (unordered, signaling) \n |
1637 | /// 0x07: Ordered (non-signaling) \n |
1638 | /// 0x08: Equal (unordered, non-signaling) \n |
1639 | /// 0x09: Not-greater-than-or-equal (unordered, signaling) \n |
1640 | /// 0x0A: Not-greater-than (unordered, signaling) \n |
1641 | /// 0x0B: False (ordered, non-signaling) \n |
1642 | /// 0x0C: Not-equal (ordered, non-signaling) \n |
1643 | /// 0x0D: Greater-than-or-equal (ordered, signaling) \n |
1644 | /// 0x0E: Greater-than (ordered, signaling) \n |
1645 | /// 0x0F: True (unordered, non-signaling) \n |
1646 | /// 0x10: Equal (ordered, signaling) \n |
1647 | /// 0x11: Less-than (ordered, non-signaling) \n |
1648 | /// 0x12: Less-than-or-equal (ordered, non-signaling) \n |
1649 | /// 0x13: Unordered (signaling) \n |
1650 | /// 0x14: Not-equal (unordered, signaling) \n |
1651 | /// 0x15: Not-less-than (unordered, non-signaling) \n |
1652 | /// 0x16: Not-less-than-or-equal (unordered, non-signaling) \n |
1653 | /// 0x17: Ordered (signaling) \n |
1654 | /// 0x18: Equal (unordered, signaling) \n |
1655 | /// 0x19: Not-greater-than-or-equal (unordered, non-signaling) \n |
1656 | /// 0x1A: Not-greater-than (unordered, non-signaling) \n |
1657 | /// 0x1B: False (ordered, signaling) \n |
1658 | /// 0x1C: Not-equal (ordered, signaling) \n |
1659 | /// 0x1D: Greater-than-or-equal (ordered, non-signaling) \n |
1660 | /// 0x1E: Greater-than (ordered, non-signaling) \n |
1661 | /// 0x1F: True (unordered, signaling) |
1662 | /// \returns A 128-bit vector of [2 x double] containing the comparison results. |
1663 | #define _mm_cmp_pd(a, b, c) \ |
1664 | (__m128d)__builtin_ia32_cmppd((__v2df)(__m128d)(a), \ |
1665 | (__v2df)(__m128d)(b), (c)) |
1666 | |
1667 | /// Compares each of the corresponding values of two 128-bit vectors of |
1668 | /// [4 x float], using the operation specified by the immediate integer |
1669 | /// operand. |
1670 | /// |
1671 | /// Returns a [4 x float] vector consisting of four floats corresponding to |
1672 | /// the four comparison results: zero if the comparison is false, and all 1's |
1673 | /// if the comparison is true. |
1674 | /// |
1675 | /// \headerfile <x86intrin.h> |
1676 | /// |
1677 | /// \code |
1678 | /// __m128 _mm_cmp_ps(__m128 a, __m128 b, const int c); |
1679 | /// \endcode |
1680 | /// |
1681 | /// This intrinsic corresponds to the <c> VCMPPS </c> instruction. |
1682 | /// |
1683 | /// \param a |
1684 | /// A 128-bit vector of [4 x float]. |
1685 | /// \param b |
1686 | /// A 128-bit vector of [4 x float]. |
1687 | /// \param c |
1688 | /// An immediate integer operand, with bits [4:0] specifying which comparison |
1689 | /// operation to use: \n |
1690 | /// 0x00: Equal (ordered, non-signaling) \n |
1691 | /// 0x01: Less-than (ordered, signaling) \n |
1692 | /// 0x02: Less-than-or-equal (ordered, signaling) \n |
1693 | /// 0x03: Unordered (non-signaling) \n |
1694 | /// 0x04: Not-equal (unordered, non-signaling) \n |
1695 | /// 0x05: Not-less-than (unordered, signaling) \n |
1696 | /// 0x06: Not-less-than-or-equal (unordered, signaling) \n |
1697 | /// 0x07: Ordered (non-signaling) \n |
1698 | /// 0x08: Equal (unordered, non-signaling) \n |
1699 | /// 0x09: Not-greater-than-or-equal (unordered, signaling) \n |
1700 | /// 0x0A: Not-greater-than (unordered, signaling) \n |
1701 | /// 0x0B: False (ordered, non-signaling) \n |
1702 | /// 0x0C: Not-equal (ordered, non-signaling) \n |
1703 | /// 0x0D: Greater-than-or-equal (ordered, signaling) \n |
1704 | /// 0x0E: Greater-than (ordered, signaling) \n |
1705 | /// 0x0F: True (unordered, non-signaling) \n |
1706 | /// 0x10: Equal (ordered, signaling) \n |
1707 | /// 0x11: Less-than (ordered, non-signaling) \n |
1708 | /// 0x12: Less-than-or-equal (ordered, non-signaling) \n |
1709 | /// 0x13: Unordered (signaling) \n |
1710 | /// 0x14: Not-equal (unordered, signaling) \n |
1711 | /// 0x15: Not-less-than (unordered, non-signaling) \n |
1712 | /// 0x16: Not-less-than-or-equal (unordered, non-signaling) \n |
1713 | /// 0x17: Ordered (signaling) \n |
1714 | /// 0x18: Equal (unordered, signaling) \n |
1715 | /// 0x19: Not-greater-than-or-equal (unordered, non-signaling) \n |
1716 | /// 0x1A: Not-greater-than (unordered, non-signaling) \n |
1717 | /// 0x1B: False (ordered, signaling) \n |
1718 | /// 0x1C: Not-equal (ordered, signaling) \n |
1719 | /// 0x1D: Greater-than-or-equal (ordered, non-signaling) \n |
1720 | /// 0x1E: Greater-than (ordered, non-signaling) \n |
1721 | /// 0x1F: True (unordered, signaling) |
1722 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
1723 | #define _mm_cmp_ps(a, b, c) \ |
1724 | (__m128)__builtin_ia32_cmpps((__v4sf)(__m128)(a), \ |
1725 | (__v4sf)(__m128)(b), (c)) |
1726 | |
1727 | /// Compares each of the corresponding double-precision values of two |
1728 | /// 256-bit vectors of [4 x double], using the operation specified by the |
1729 | /// immediate integer operand. |
1730 | /// |
1731 | /// Returns a [4 x double] vector consisting of four doubles corresponding to |
1732 | /// the four comparison results: zero if the comparison is false, and all 1's |
1733 | /// if the comparison is true. |
1734 | /// |
1735 | /// \headerfile <x86intrin.h> |
1736 | /// |
1737 | /// \code |
1738 | /// __m256d _mm256_cmp_pd(__m256d a, __m256d b, const int c); |
1739 | /// \endcode |
1740 | /// |
1741 | /// This intrinsic corresponds to the <c> VCMPPD </c> instruction. |
1742 | /// |
1743 | /// \param a |
1744 | /// A 256-bit vector of [4 x double]. |
1745 | /// \param b |
1746 | /// A 256-bit vector of [4 x double]. |
1747 | /// \param c |
1748 | /// An immediate integer operand, with bits [4:0] specifying which comparison |
1749 | /// operation to use: \n |
1750 | /// 0x00: Equal (ordered, non-signaling) \n |
1751 | /// 0x01: Less-than (ordered, signaling) \n |
1752 | /// 0x02: Less-than-or-equal (ordered, signaling) \n |
1753 | /// 0x03: Unordered (non-signaling) \n |
1754 | /// 0x04: Not-equal (unordered, non-signaling) \n |
1755 | /// 0x05: Not-less-than (unordered, signaling) \n |
1756 | /// 0x06: Not-less-than-or-equal (unordered, signaling) \n |
1757 | /// 0x07: Ordered (non-signaling) \n |
1758 | /// 0x08: Equal (unordered, non-signaling) \n |
1759 | /// 0x09: Not-greater-than-or-equal (unordered, signaling) \n |
1760 | /// 0x0A: Not-greater-than (unordered, signaling) \n |
1761 | /// 0x0B: False (ordered, non-signaling) \n |
1762 | /// 0x0C: Not-equal (ordered, non-signaling) \n |
1763 | /// 0x0D: Greater-than-or-equal (ordered, signaling) \n |
1764 | /// 0x0E: Greater-than (ordered, signaling) \n |
1765 | /// 0x0F: True (unordered, non-signaling) \n |
1766 | /// 0x10: Equal (ordered, signaling) \n |
1767 | /// 0x11: Less-than (ordered, non-signaling) \n |
1768 | /// 0x12: Less-than-or-equal (ordered, non-signaling) \n |
1769 | /// 0x13: Unordered (signaling) \n |
1770 | /// 0x14: Not-equal (unordered, signaling) \n |
1771 | /// 0x15: Not-less-than (unordered, non-signaling) \n |
1772 | /// 0x16: Not-less-than-or-equal (unordered, non-signaling) \n |
1773 | /// 0x17: Ordered (signaling) \n |
1774 | /// 0x18: Equal (unordered, signaling) \n |
1775 | /// 0x19: Not-greater-than-or-equal (unordered, non-signaling) \n |
1776 | /// 0x1A: Not-greater-than (unordered, non-signaling) \n |
1777 | /// 0x1B: False (ordered, signaling) \n |
1778 | /// 0x1C: Not-equal (ordered, signaling) \n |
1779 | /// 0x1D: Greater-than-or-equal (ordered, non-signaling) \n |
1780 | /// 0x1E: Greater-than (ordered, non-signaling) \n |
1781 | /// 0x1F: True (unordered, signaling) |
1782 | /// \returns A 256-bit vector of [4 x double] containing the comparison results. |
1783 | #define _mm256_cmp_pd(a, b, c) \ |
1784 | (__m256d)__builtin_ia32_cmppd256((__v4df)(__m256d)(a), \ |
1785 | (__v4df)(__m256d)(b), (c)) |
1786 | |
1787 | /// Compares each of the corresponding values of two 256-bit vectors of |
1788 | /// [8 x float], using the operation specified by the immediate integer |
1789 | /// operand. |
1790 | /// |
1791 | /// Returns a [8 x float] vector consisting of eight floats corresponding to |
1792 | /// the eight comparison results: zero if the comparison is false, and all |
1793 | /// 1's if the comparison is true. |
1794 | /// |
1795 | /// \headerfile <x86intrin.h> |
1796 | /// |
1797 | /// \code |
1798 | /// __m256 _mm256_cmp_ps(__m256 a, __m256 b, const int c); |
1799 | /// \endcode |
1800 | /// |
1801 | /// This intrinsic corresponds to the <c> VCMPPS </c> instruction. |
1802 | /// |
1803 | /// \param a |
1804 | /// A 256-bit vector of [8 x float]. |
1805 | /// \param b |
1806 | /// A 256-bit vector of [8 x float]. |
1807 | /// \param c |
1808 | /// An immediate integer operand, with bits [4:0] specifying which comparison |
1809 | /// operation to use: \n |
1810 | /// 0x00: Equal (ordered, non-signaling) \n |
1811 | /// 0x01: Less-than (ordered, signaling) \n |
1812 | /// 0x02: Less-than-or-equal (ordered, signaling) \n |
1813 | /// 0x03: Unordered (non-signaling) \n |
1814 | /// 0x04: Not-equal (unordered, non-signaling) \n |
1815 | /// 0x05: Not-less-than (unordered, signaling) \n |
1816 | /// 0x06: Not-less-than-or-equal (unordered, signaling) \n |
1817 | /// 0x07: Ordered (non-signaling) \n |
1818 | /// 0x08: Equal (unordered, non-signaling) \n |
1819 | /// 0x09: Not-greater-than-or-equal (unordered, signaling) \n |
1820 | /// 0x0A: Not-greater-than (unordered, signaling) \n |
1821 | /// 0x0B: False (ordered, non-signaling) \n |
1822 | /// 0x0C: Not-equal (ordered, non-signaling) \n |
1823 | /// 0x0D: Greater-than-or-equal (ordered, signaling) \n |
1824 | /// 0x0E: Greater-than (ordered, signaling) \n |
1825 | /// 0x0F: True (unordered, non-signaling) \n |
1826 | /// 0x10: Equal (ordered, signaling) \n |
1827 | /// 0x11: Less-than (ordered, non-signaling) \n |
1828 | /// 0x12: Less-than-or-equal (ordered, non-signaling) \n |
1829 | /// 0x13: Unordered (signaling) \n |
1830 | /// 0x14: Not-equal (unordered, signaling) \n |
1831 | /// 0x15: Not-less-than (unordered, non-signaling) \n |
1832 | /// 0x16: Not-less-than-or-equal (unordered, non-signaling) \n |
1833 | /// 0x17: Ordered (signaling) \n |
1834 | /// 0x18: Equal (unordered, signaling) \n |
1835 | /// 0x19: Not-greater-than-or-equal (unordered, non-signaling) \n |
1836 | /// 0x1A: Not-greater-than (unordered, non-signaling) \n |
1837 | /// 0x1B: False (ordered, signaling) \n |
1838 | /// 0x1C: Not-equal (ordered, signaling) \n |
1839 | /// 0x1D: Greater-than-or-equal (ordered, non-signaling) \n |
1840 | /// 0x1E: Greater-than (ordered, non-signaling) \n |
1841 | /// 0x1F: True (unordered, signaling) |
1842 | /// \returns A 256-bit vector of [8 x float] containing the comparison results. |
1843 | #define _mm256_cmp_ps(a, b, c) \ |
1844 | (__m256)__builtin_ia32_cmpps256((__v8sf)(__m256)(a), \ |
1845 | (__v8sf)(__m256)(b), (c)) |
1846 | |
1847 | /// Compares each of the corresponding scalar double-precision values of |
1848 | /// two 128-bit vectors of [2 x double], using the operation specified by the |
1849 | /// immediate integer operand. |
1850 | /// |
1851 | /// If the result is true, all 64 bits of the destination vector are set; |
1852 | /// otherwise they are cleared. |
1853 | /// |
1854 | /// \headerfile <x86intrin.h> |
1855 | /// |
1856 | /// \code |
1857 | /// __m128d _mm_cmp_sd(__m128d a, __m128d b, const int c); |
1858 | /// \endcode |
1859 | /// |
1860 | /// This intrinsic corresponds to the <c> VCMPSD </c> instruction. |
1861 | /// |
1862 | /// \param a |
1863 | /// A 128-bit vector of [2 x double]. |
1864 | /// \param b |
1865 | /// A 128-bit vector of [2 x double]. |
1866 | /// \param c |
1867 | /// An immediate integer operand, with bits [4:0] specifying which comparison |
1868 | /// operation to use: \n |
1869 | /// 0x00: Equal (ordered, non-signaling) \n |
1870 | /// 0x01: Less-than (ordered, signaling) \n |
1871 | /// 0x02: Less-than-or-equal (ordered, signaling) \n |
1872 | /// 0x03: Unordered (non-signaling) \n |
1873 | /// 0x04: Not-equal (unordered, non-signaling) \n |
1874 | /// 0x05: Not-less-than (unordered, signaling) \n |
1875 | /// 0x06: Not-less-than-or-equal (unordered, signaling) \n |
1876 | /// 0x07: Ordered (non-signaling) \n |
1877 | /// 0x08: Equal (unordered, non-signaling) \n |
1878 | /// 0x09: Not-greater-than-or-equal (unordered, signaling) \n |
1879 | /// 0x0A: Not-greater-than (unordered, signaling) \n |
1880 | /// 0x0B: False (ordered, non-signaling) \n |
1881 | /// 0x0C: Not-equal (ordered, non-signaling) \n |
1882 | /// 0x0D: Greater-than-or-equal (ordered, signaling) \n |
1883 | /// 0x0E: Greater-than (ordered, signaling) \n |
1884 | /// 0x0F: True (unordered, non-signaling) \n |
1885 | /// 0x10: Equal (ordered, signaling) \n |
1886 | /// 0x11: Less-than (ordered, non-signaling) \n |
1887 | /// 0x12: Less-than-or-equal (ordered, non-signaling) \n |
1888 | /// 0x13: Unordered (signaling) \n |
1889 | /// 0x14: Not-equal (unordered, signaling) \n |
1890 | /// 0x15: Not-less-than (unordered, non-signaling) \n |
1891 | /// 0x16: Not-less-than-or-equal (unordered, non-signaling) \n |
1892 | /// 0x17: Ordered (signaling) \n |
1893 | /// 0x18: Equal (unordered, signaling) \n |
1894 | /// 0x19: Not-greater-than-or-equal (unordered, non-signaling) \n |
1895 | /// 0x1A: Not-greater-than (unordered, non-signaling) \n |
1896 | /// 0x1B: False (ordered, signaling) \n |
1897 | /// 0x1C: Not-equal (ordered, signaling) \n |
1898 | /// 0x1D: Greater-than-or-equal (ordered, non-signaling) \n |
1899 | /// 0x1E: Greater-than (ordered, non-signaling) \n |
1900 | /// 0x1F: True (unordered, signaling) |
1901 | /// \returns A 128-bit vector of [2 x double] containing the comparison results. |
1902 | #define _mm_cmp_sd(a, b, c) \ |
1903 | (__m128d)__builtin_ia32_cmpsd((__v2df)(__m128d)(a), \ |
1904 | (__v2df)(__m128d)(b), (c)) |
1905 | |
1906 | /// Compares each of the corresponding scalar values of two 128-bit |
1907 | /// vectors of [4 x float], using the operation specified by the immediate |
1908 | /// integer operand. |
1909 | /// |
1910 | /// If the result is true, all 32 bits of the destination vector are set; |
1911 | /// otherwise they are cleared. |
1912 | /// |
1913 | /// \headerfile <x86intrin.h> |
1914 | /// |
1915 | /// \code |
1916 | /// __m128 _mm_cmp_ss(__m128 a, __m128 b, const int c); |
1917 | /// \endcode |
1918 | /// |
1919 | /// This intrinsic corresponds to the <c> VCMPSS </c> instruction. |
1920 | /// |
1921 | /// \param a |
1922 | /// A 128-bit vector of [4 x float]. |
1923 | /// \param b |
1924 | /// A 128-bit vector of [4 x float]. |
1925 | /// \param c |
1926 | /// An immediate integer operand, with bits [4:0] specifying which comparison |
1927 | /// operation to use: \n |
1928 | /// 0x00: Equal (ordered, non-signaling) \n |
1929 | /// 0x01: Less-than (ordered, signaling) \n |
1930 | /// 0x02: Less-than-or-equal (ordered, signaling) \n |
1931 | /// 0x03: Unordered (non-signaling) \n |
1932 | /// 0x04: Not-equal (unordered, non-signaling) \n |
1933 | /// 0x05: Not-less-than (unordered, signaling) \n |
1934 | /// 0x06: Not-less-than-or-equal (unordered, signaling) \n |
1935 | /// 0x07: Ordered (non-signaling) \n |
1936 | /// 0x08: Equal (unordered, non-signaling) \n |
1937 | /// 0x09: Not-greater-than-or-equal (unordered, signaling) \n |
1938 | /// 0x0A: Not-greater-than (unordered, signaling) \n |
1939 | /// 0x0B: False (ordered, non-signaling) \n |
1940 | /// 0x0C: Not-equal (ordered, non-signaling) \n |
1941 | /// 0x0D: Greater-than-or-equal (ordered, signaling) \n |
1942 | /// 0x0E: Greater-than (ordered, signaling) \n |
1943 | /// 0x0F: True (unordered, non-signaling) \n |
1944 | /// 0x10: Equal (ordered, signaling) \n |
1945 | /// 0x11: Less-than (ordered, non-signaling) \n |
1946 | /// 0x12: Less-than-or-equal (ordered, non-signaling) \n |
1947 | /// 0x13: Unordered (signaling) \n |
1948 | /// 0x14: Not-equal (unordered, signaling) \n |
1949 | /// 0x15: Not-less-than (unordered, non-signaling) \n |
1950 | /// 0x16: Not-less-than-or-equal (unordered, non-signaling) \n |
1951 | /// 0x17: Ordered (signaling) \n |
1952 | /// 0x18: Equal (unordered, signaling) \n |
1953 | /// 0x19: Not-greater-than-or-equal (unordered, non-signaling) \n |
1954 | /// 0x1A: Not-greater-than (unordered, non-signaling) \n |
1955 | /// 0x1B: False (ordered, signaling) \n |
1956 | /// 0x1C: Not-equal (ordered, signaling) \n |
1957 | /// 0x1D: Greater-than-or-equal (ordered, non-signaling) \n |
1958 | /// 0x1E: Greater-than (ordered, non-signaling) \n |
1959 | /// 0x1F: True (unordered, signaling) |
1960 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
1961 | #define _mm_cmp_ss(a, b, c) \ |
1962 | (__m128)__builtin_ia32_cmpss((__v4sf)(__m128)(a), \ |
1963 | (__v4sf)(__m128)(b), (c)) |
1964 | |
1965 | /// Takes a [8 x i32] vector and returns the vector element value |
1966 | /// indexed by the immediate constant operand. |
1967 | /// |
1968 | /// \headerfile <x86intrin.h> |
1969 | /// |
1970 | /// This intrinsic corresponds to the <c> VEXTRACTF128+COMPOSITE </c> |
1971 | /// instruction. |
1972 | /// |
1973 | /// \param __a |
1974 | /// A 256-bit vector of [8 x i32]. |
1975 | /// \param __imm |
1976 | /// An immediate integer operand with bits [2:0] determining which vector |
1977 | /// element is extracted and returned. |
1978 | /// \returns A 32-bit integer containing the extracted 32 bits of extended |
1979 | /// packed data. |
1980 | #define _mm256_extract_epi32(X, N) \ |
1981 | (int)__builtin_ia32_vec_ext_v8si((__v8si)(__m256i)(X), (int)(N)) |
1982 | |
1983 | /// Takes a [16 x i16] vector and returns the vector element value |
1984 | /// indexed by the immediate constant operand. |
1985 | /// |
1986 | /// \headerfile <x86intrin.h> |
1987 | /// |
1988 | /// This intrinsic corresponds to the <c> VEXTRACTF128+COMPOSITE </c> |
1989 | /// instruction. |
1990 | /// |
1991 | /// \param __a |
1992 | /// A 256-bit integer vector of [16 x i16]. |
1993 | /// \param __imm |
1994 | /// An immediate integer operand with bits [3:0] determining which vector |
1995 | /// element is extracted and returned. |
1996 | /// \returns A 32-bit integer containing the extracted 16 bits of zero extended |
1997 | /// packed data. |
1998 | #define _mm256_extract_epi16(X, N) \ |
1999 | (int)(unsigned short)__builtin_ia32_vec_ext_v16hi((__v16hi)(__m256i)(X), \ |
2000 | (int)(N)) |
2001 | |
2002 | /// Takes a [32 x i8] vector and returns the vector element value |
2003 | /// indexed by the immediate constant operand. |
2004 | /// |
2005 | /// \headerfile <x86intrin.h> |
2006 | /// |
2007 | /// This intrinsic corresponds to the <c> VEXTRACTF128+COMPOSITE </c> |
2008 | /// instruction. |
2009 | /// |
2010 | /// \param __a |
2011 | /// A 256-bit integer vector of [32 x i8]. |
2012 | /// \param __imm |
2013 | /// An immediate integer operand with bits [4:0] determining which vector |
2014 | /// element is extracted and returned. |
2015 | /// \returns A 32-bit integer containing the extracted 8 bits of zero extended |
2016 | /// packed data. |
2017 | #define _mm256_extract_epi8(X, N) \ |
2018 | (int)(unsigned char)__builtin_ia32_vec_ext_v32qi((__v32qi)(__m256i)(X), \ |
2019 | (int)(N)) |
2020 | |
2021 | #ifdef __x86_64__ |
2022 | /// Takes a [4 x i64] vector and returns the vector element value |
2023 | /// indexed by the immediate constant operand. |
2024 | /// |
2025 | /// \headerfile <x86intrin.h> |
2026 | /// |
2027 | /// This intrinsic corresponds to the <c> VEXTRACTF128+COMPOSITE </c> |
2028 | /// instruction. |
2029 | /// |
2030 | /// \param __a |
2031 | /// A 256-bit integer vector of [4 x i64]. |
2032 | /// \param __imm |
2033 | /// An immediate integer operand with bits [1:0] determining which vector |
2034 | /// element is extracted and returned. |
2035 | /// \returns A 64-bit integer containing the extracted 64 bits of extended |
2036 | /// packed data. |
2037 | #define _mm256_extract_epi64(X, N) \ |
2038 | (long long)__builtin_ia32_vec_ext_v4di((__v4di)(__m256i)(X), (int)(N)) |
2039 | #endif |
2040 | |
2041 | /// Takes a [8 x i32] vector and replaces the vector element value |
2042 | /// indexed by the immediate constant operand by a new value. Returns the |
2043 | /// modified vector. |
2044 | /// |
2045 | /// \headerfile <x86intrin.h> |
2046 | /// |
2047 | /// This intrinsic corresponds to the <c> VINSERTF128+COMPOSITE </c> |
2048 | /// instruction. |
2049 | /// |
2050 | /// \param __a |
2051 | /// A vector of [8 x i32] to be used by the insert operation. |
2052 | /// \param __b |
2053 | /// An integer value. The replacement value for the insert operation. |
2054 | /// \param __imm |
2055 | /// An immediate integer specifying the index of the vector element to be |
2056 | /// replaced. |
2057 | /// \returns A copy of vector \a __a, after replacing its element indexed by |
2058 | /// \a __imm with \a __b. |
2059 | #define _mm256_insert_epi32(X, I, N) \ |
2060 | (__m256i)__builtin_ia32_vec_set_v8si((__v8si)(__m256i)(X), \ |
2061 | (int)(I), (int)(N)) |
2062 | |
2063 | |
2064 | /// Takes a [16 x i16] vector and replaces the vector element value |
2065 | /// indexed by the immediate constant operand with a new value. Returns the |
2066 | /// modified vector. |
2067 | /// |
2068 | /// \headerfile <x86intrin.h> |
2069 | /// |
2070 | /// This intrinsic corresponds to the <c> VINSERTF128+COMPOSITE </c> |
2071 | /// instruction. |
2072 | /// |
2073 | /// \param __a |
2074 | /// A vector of [16 x i16] to be used by the insert operation. |
2075 | /// \param __b |
2076 | /// An i16 integer value. The replacement value for the insert operation. |
2077 | /// \param __imm |
2078 | /// An immediate integer specifying the index of the vector element to be |
2079 | /// replaced. |
2080 | /// \returns A copy of vector \a __a, after replacing its element indexed by |
2081 | /// \a __imm with \a __b. |
2082 | #define _mm256_insert_epi16(X, I, N) \ |
2083 | (__m256i)__builtin_ia32_vec_set_v16hi((__v16hi)(__m256i)(X), \ |
2084 | (int)(I), (int)(N)) |
2085 | |
2086 | /// Takes a [32 x i8] vector and replaces the vector element value |
2087 | /// indexed by the immediate constant operand with a new value. Returns the |
2088 | /// modified vector. |
2089 | /// |
2090 | /// \headerfile <x86intrin.h> |
2091 | /// |
2092 | /// This intrinsic corresponds to the <c> VINSERTF128+COMPOSITE </c> |
2093 | /// instruction. |
2094 | /// |
2095 | /// \param __a |
2096 | /// A vector of [32 x i8] to be used by the insert operation. |
2097 | /// \param __b |
2098 | /// An i8 integer value. The replacement value for the insert operation. |
2099 | /// \param __imm |
2100 | /// An immediate integer specifying the index of the vector element to be |
2101 | /// replaced. |
2102 | /// \returns A copy of vector \a __a, after replacing its element indexed by |
2103 | /// \a __imm with \a __b. |
2104 | #define _mm256_insert_epi8(X, I, N) \ |
2105 | (__m256i)__builtin_ia32_vec_set_v32qi((__v32qi)(__m256i)(X), \ |
2106 | (int)(I), (int)(N)) |
2107 | |
2108 | #ifdef __x86_64__ |
2109 | /// Takes a [4 x i64] vector and replaces the vector element value |
2110 | /// indexed by the immediate constant operand with a new value. Returns the |
2111 | /// modified vector. |
2112 | /// |
2113 | /// \headerfile <x86intrin.h> |
2114 | /// |
2115 | /// This intrinsic corresponds to the <c> VINSERTF128+COMPOSITE </c> |
2116 | /// instruction. |
2117 | /// |
2118 | /// \param __a |
2119 | /// A vector of [4 x i64] to be used by the insert operation. |
2120 | /// \param __b |
2121 | /// A 64-bit integer value. The replacement value for the insert operation. |
2122 | /// \param __imm |
2123 | /// An immediate integer specifying the index of the vector element to be |
2124 | /// replaced. |
2125 | /// \returns A copy of vector \a __a, after replacing its element indexed by |
2126 | /// \a __imm with \a __b. |
2127 | #define _mm256_insert_epi64(X, I, N) \ |
2128 | (__m256i)__builtin_ia32_vec_set_v4di((__v4di)(__m256i)(X), \ |
2129 | (long long)(I), (int)(N)) |
2130 | #endif |
2131 | |
2132 | /* Conversion */ |
2133 | /// Converts a vector of [4 x i32] into a vector of [4 x double]. |
2134 | /// |
2135 | /// \headerfile <x86intrin.h> |
2136 | /// |
2137 | /// This intrinsic corresponds to the <c> VCVTDQ2PD </c> instruction. |
2138 | /// |
2139 | /// \param __a |
2140 | /// A 128-bit integer vector of [4 x i32]. |
2141 | /// \returns A 256-bit vector of [4 x double] containing the converted values. |
2142 | static __inline __m256d __DEFAULT_FN_ATTRS |
2143 | _mm256_cvtepi32_pd(__m128i __a) |
2144 | { |
2145 | return (__m256d)__builtin_convertvector((__v4si)__a, __v4df); |
2146 | } |
2147 | |
2148 | /// Converts a vector of [8 x i32] into a vector of [8 x float]. |
2149 | /// |
2150 | /// \headerfile <x86intrin.h> |
2151 | /// |
2152 | /// This intrinsic corresponds to the <c> VCVTDQ2PS </c> instruction. |
2153 | /// |
2154 | /// \param __a |
2155 | /// A 256-bit integer vector. |
2156 | /// \returns A 256-bit vector of [8 x float] containing the converted values. |
2157 | static __inline __m256 __DEFAULT_FN_ATTRS |
2158 | _mm256_cvtepi32_ps(__m256i __a) |
2159 | { |
2160 | return (__m256)__builtin_convertvector((__v8si)__a, __v8sf); |
2161 | } |
2162 | |
2163 | /// Converts a 256-bit vector of [4 x double] into a 128-bit vector of |
2164 | /// [4 x float]. |
2165 | /// |
2166 | /// \headerfile <x86intrin.h> |
2167 | /// |
2168 | /// This intrinsic corresponds to the <c> VCVTPD2PS </c> instruction. |
2169 | /// |
2170 | /// \param __a |
2171 | /// A 256-bit vector of [4 x double]. |
2172 | /// \returns A 128-bit vector of [4 x float] containing the converted values. |
2173 | static __inline __m128 __DEFAULT_FN_ATTRS |
2174 | _mm256_cvtpd_ps(__m256d __a) |
2175 | { |
2176 | return (__m128)__builtin_ia32_cvtpd2ps256((__v4df) __a); |
2177 | } |
2178 | |
2179 | /// Converts a vector of [8 x float] into a vector of [8 x i32]. |
2180 | /// |
2181 | /// \headerfile <x86intrin.h> |
2182 | /// |
2183 | /// This intrinsic corresponds to the <c> VCVTPS2DQ </c> instruction. |
2184 | /// |
2185 | /// \param __a |
2186 | /// A 256-bit vector of [8 x float]. |
2187 | /// \returns A 256-bit integer vector containing the converted values. |
2188 | static __inline __m256i __DEFAULT_FN_ATTRS |
2189 | _mm256_cvtps_epi32(__m256 __a) |
2190 | { |
2191 | return (__m256i)__builtin_ia32_cvtps2dq256((__v8sf) __a); |
2192 | } |
2193 | |
2194 | /// Converts a 128-bit vector of [4 x float] into a 256-bit vector of [4 |
2195 | /// x double]. |
2196 | /// |
2197 | /// \headerfile <x86intrin.h> |
2198 | /// |
2199 | /// This intrinsic corresponds to the <c> VCVTPS2PD </c> instruction. |
2200 | /// |
2201 | /// \param __a |
2202 | /// A 128-bit vector of [4 x float]. |
2203 | /// \returns A 256-bit vector of [4 x double] containing the converted values. |
2204 | static __inline __m256d __DEFAULT_FN_ATTRS |
2205 | _mm256_cvtps_pd(__m128 __a) |
2206 | { |
2207 | return (__m256d)__builtin_convertvector((__v4sf)__a, __v4df); |
2208 | } |
2209 | |
2210 | /// Converts a 256-bit vector of [4 x double] into a 128-bit vector of [4 |
2211 | /// x i32], truncating the result by rounding towards zero when it is |
2212 | /// inexact. |
2213 | /// |
2214 | /// \headerfile <x86intrin.h> |
2215 | /// |
2216 | /// This intrinsic corresponds to the <c> VCVTTPD2DQ </c> instruction. |
2217 | /// |
2218 | /// \param __a |
2219 | /// A 256-bit vector of [4 x double]. |
2220 | /// \returns A 128-bit integer vector containing the converted values. |
2221 | static __inline __m128i __DEFAULT_FN_ATTRS |
2222 | _mm256_cvttpd_epi32(__m256d __a) |
2223 | { |
2224 | return (__m128i)__builtin_ia32_cvttpd2dq256((__v4df) __a); |
2225 | } |
2226 | |
2227 | /// Converts a 256-bit vector of [4 x double] into a 128-bit vector of [4 |
2228 | /// x i32]. When a conversion is inexact, the value returned is rounded |
2229 | /// according to the rounding control bits in the MXCSR register. |
2230 | /// |
2231 | /// \headerfile <x86intrin.h> |
2232 | /// |
2233 | /// This intrinsic corresponds to the <c> VCVTPD2DQ </c> instruction. |
2234 | /// |
2235 | /// \param __a |
2236 | /// A 256-bit vector of [4 x double]. |
2237 | /// \returns A 128-bit integer vector containing the converted values. |
2238 | static __inline __m128i __DEFAULT_FN_ATTRS |
2239 | _mm256_cvtpd_epi32(__m256d __a) |
2240 | { |
2241 | return (__m128i)__builtin_ia32_cvtpd2dq256((__v4df) __a); |
2242 | } |
2243 | |
2244 | /// Converts a vector of [8 x float] into a vector of [8 x i32], |
2245 | /// truncating the result by rounding towards zero when it is inexact. |
2246 | /// |
2247 | /// \headerfile <x86intrin.h> |
2248 | /// |
2249 | /// This intrinsic corresponds to the <c> VCVTTPS2DQ </c> instruction. |
2250 | /// |
2251 | /// \param __a |
2252 | /// A 256-bit vector of [8 x float]. |
2253 | /// \returns A 256-bit integer vector containing the converted values. |
2254 | static __inline __m256i __DEFAULT_FN_ATTRS |
2255 | _mm256_cvttps_epi32(__m256 __a) |
2256 | { |
2257 | return (__m256i)__builtin_ia32_cvttps2dq256((__v8sf) __a); |
2258 | } |
2259 | |
2260 | /// Returns the first element of the input vector of [4 x double]. |
2261 | /// |
2262 | /// \headerfile <avxintrin.h> |
2263 | /// |
2264 | /// This intrinsic is a utility function and does not correspond to a specific |
2265 | /// instruction. |
2266 | /// |
2267 | /// \param __a |
2268 | /// A 256-bit vector of [4 x double]. |
2269 | /// \returns A 64 bit double containing the first element of the input vector. |
2270 | static __inline double __DEFAULT_FN_ATTRS |
2271 | _mm256_cvtsd_f64(__m256d __a) |
2272 | { |
2273 | return __a[0]; |
2274 | } |
2275 | |
2276 | /// Returns the first element of the input vector of [8 x i32]. |
2277 | /// |
2278 | /// \headerfile <avxintrin.h> |
2279 | /// |
2280 | /// This intrinsic is a utility function and does not correspond to a specific |
2281 | /// instruction. |
2282 | /// |
2283 | /// \param __a |
2284 | /// A 256-bit vector of [8 x i32]. |
2285 | /// \returns A 32 bit integer containing the first element of the input vector. |
2286 | static __inline int __DEFAULT_FN_ATTRS |
2287 | _mm256_cvtsi256_si32(__m256i __a) |
2288 | { |
2289 | __v8si __b = (__v8si)__a; |
2290 | return __b[0]; |
2291 | } |
2292 | |
2293 | /// Returns the first element of the input vector of [8 x float]. |
2294 | /// |
2295 | /// \headerfile <avxintrin.h> |
2296 | /// |
2297 | /// This intrinsic is a utility function and does not correspond to a specific |
2298 | /// instruction. |
2299 | /// |
2300 | /// \param __a |
2301 | /// A 256-bit vector of [8 x float]. |
2302 | /// \returns A 32 bit float containing the first element of the input vector. |
2303 | static __inline float __DEFAULT_FN_ATTRS |
2304 | _mm256_cvtss_f32(__m256 __a) |
2305 | { |
2306 | return __a[0]; |
2307 | } |
2308 | |
2309 | /* Vector replicate */ |
2310 | /// Moves and duplicates odd-indexed values from a 256-bit vector of |
2311 | /// [8 x float] to float values in a 256-bit vector of [8 x float]. |
2312 | /// |
2313 | /// \headerfile <x86intrin.h> |
2314 | /// |
2315 | /// This intrinsic corresponds to the <c> VMOVSHDUP </c> instruction. |
2316 | /// |
2317 | /// \param __a |
2318 | /// A 256-bit vector of [8 x float]. \n |
2319 | /// Bits [255:224] of \a __a are written to bits [255:224] and [223:192] of |
2320 | /// the return value. \n |
2321 | /// Bits [191:160] of \a __a are written to bits [191:160] and [159:128] of |
2322 | /// the return value. \n |
2323 | /// Bits [127:96] of \a __a are written to bits [127:96] and [95:64] of the |
2324 | /// return value. \n |
2325 | /// Bits [63:32] of \a __a are written to bits [63:32] and [31:0] of the |
2326 | /// return value. |
2327 | /// \returns A 256-bit vector of [8 x float] containing the moved and duplicated |
2328 | /// values. |
2329 | static __inline __m256 __DEFAULT_FN_ATTRS |
2330 | _mm256_movehdup_ps(__m256 __a) |
2331 | { |
2332 | return __builtin_shufflevector((__v8sf)__a, (__v8sf)__a, 1, 1, 3, 3, 5, 5, 7, 7); |
2333 | } |
2334 | |
2335 | /// Moves and duplicates even-indexed values from a 256-bit vector of |
2336 | /// [8 x float] to float values in a 256-bit vector of [8 x float]. |
2337 | /// |
2338 | /// \headerfile <x86intrin.h> |
2339 | /// |
2340 | /// This intrinsic corresponds to the <c> VMOVSLDUP </c> instruction. |
2341 | /// |
2342 | /// \param __a |
2343 | /// A 256-bit vector of [8 x float]. \n |
2344 | /// Bits [223:192] of \a __a are written to bits [255:224] and [223:192] of |
2345 | /// the return value. \n |
2346 | /// Bits [159:128] of \a __a are written to bits [191:160] and [159:128] of |
2347 | /// the return value. \n |
2348 | /// Bits [95:64] of \a __a are written to bits [127:96] and [95:64] of the |
2349 | /// return value. \n |
2350 | /// Bits [31:0] of \a __a are written to bits [63:32] and [31:0] of the |
2351 | /// return value. |
2352 | /// \returns A 256-bit vector of [8 x float] containing the moved and duplicated |
2353 | /// values. |
2354 | static __inline __m256 __DEFAULT_FN_ATTRS |
2355 | _mm256_moveldup_ps(__m256 __a) |
2356 | { |
2357 | return __builtin_shufflevector((__v8sf)__a, (__v8sf)__a, 0, 0, 2, 2, 4, 4, 6, 6); |
2358 | } |
2359 | |
2360 | /// Moves and duplicates double-precision floating point values from a |
2361 | /// 256-bit vector of [4 x double] to double-precision values in a 256-bit |
2362 | /// vector of [4 x double]. |
2363 | /// |
2364 | /// \headerfile <x86intrin.h> |
2365 | /// |
2366 | /// This intrinsic corresponds to the <c> VMOVDDUP </c> instruction. |
2367 | /// |
2368 | /// \param __a |
2369 | /// A 256-bit vector of [4 x double]. \n |
2370 | /// Bits [63:0] of \a __a are written to bits [127:64] and [63:0] of the |
2371 | /// return value. \n |
2372 | /// Bits [191:128] of \a __a are written to bits [255:192] and [191:128] of |
2373 | /// the return value. |
2374 | /// \returns A 256-bit vector of [4 x double] containing the moved and |
2375 | /// duplicated values. |
2376 | static __inline __m256d __DEFAULT_FN_ATTRS |
2377 | _mm256_movedup_pd(__m256d __a) |
2378 | { |
2379 | return __builtin_shufflevector((__v4df)__a, (__v4df)__a, 0, 0, 2, 2); |
2380 | } |
2381 | |
2382 | /* Unpack and Interleave */ |
2383 | /// Unpacks the odd-indexed vector elements from two 256-bit vectors of |
2384 | /// [4 x double] and interleaves them into a 256-bit vector of [4 x double]. |
2385 | /// |
2386 | /// \headerfile <x86intrin.h> |
2387 | /// |
2388 | /// This intrinsic corresponds to the <c> VUNPCKHPD </c> instruction. |
2389 | /// |
2390 | /// \param __a |
2391 | /// A 256-bit floating-point vector of [4 x double]. \n |
2392 | /// Bits [127:64] are written to bits [63:0] of the return value. \n |
2393 | /// Bits [255:192] are written to bits [191:128] of the return value. \n |
2394 | /// \param __b |
2395 | /// A 256-bit floating-point vector of [4 x double]. \n |
2396 | /// Bits [127:64] are written to bits [127:64] of the return value. \n |
2397 | /// Bits [255:192] are written to bits [255:192] of the return value. \n |
2398 | /// \returns A 256-bit vector of [4 x double] containing the interleaved values. |
2399 | static __inline __m256d __DEFAULT_FN_ATTRS |
2400 | _mm256_unpackhi_pd(__m256d __a, __m256d __b) |
2401 | { |
2402 | return __builtin_shufflevector((__v4df)__a, (__v4df)__b, 1, 5, 1+2, 5+2); |
2403 | } |
2404 | |
2405 | /// Unpacks the even-indexed vector elements from two 256-bit vectors of |
2406 | /// [4 x double] and interleaves them into a 256-bit vector of [4 x double]. |
2407 | /// |
2408 | /// \headerfile <x86intrin.h> |
2409 | /// |
2410 | /// This intrinsic corresponds to the <c> VUNPCKLPD </c> instruction. |
2411 | /// |
2412 | /// \param __a |
2413 | /// A 256-bit floating-point vector of [4 x double]. \n |
2414 | /// Bits [63:0] are written to bits [63:0] of the return value. \n |
2415 | /// Bits [191:128] are written to bits [191:128] of the return value. |
2416 | /// \param __b |
2417 | /// A 256-bit floating-point vector of [4 x double]. \n |
2418 | /// Bits [63:0] are written to bits [127:64] of the return value. \n |
2419 | /// Bits [191:128] are written to bits [255:192] of the return value. \n |
2420 | /// \returns A 256-bit vector of [4 x double] containing the interleaved values. |
2421 | static __inline __m256d __DEFAULT_FN_ATTRS |
2422 | _mm256_unpacklo_pd(__m256d __a, __m256d __b) |
2423 | { |
2424 | return __builtin_shufflevector((__v4df)__a, (__v4df)__b, 0, 4, 0+2, 4+2); |
2425 | } |
2426 | |
2427 | /// Unpacks the 32-bit vector elements 2, 3, 6 and 7 from each of the |
2428 | /// two 256-bit vectors of [8 x float] and interleaves them into a 256-bit |
2429 | /// vector of [8 x float]. |
2430 | /// |
2431 | /// \headerfile <x86intrin.h> |
2432 | /// |
2433 | /// This intrinsic corresponds to the <c> VUNPCKHPS </c> instruction. |
2434 | /// |
2435 | /// \param __a |
2436 | /// A 256-bit vector of [8 x float]. \n |
2437 | /// Bits [95:64] are written to bits [31:0] of the return value. \n |
2438 | /// Bits [127:96] are written to bits [95:64] of the return value. \n |
2439 | /// Bits [223:192] are written to bits [159:128] of the return value. \n |
2440 | /// Bits [255:224] are written to bits [223:192] of the return value. |
2441 | /// \param __b |
2442 | /// A 256-bit vector of [8 x float]. \n |
2443 | /// Bits [95:64] are written to bits [63:32] of the return value. \n |
2444 | /// Bits [127:96] are written to bits [127:96] of the return value. \n |
2445 | /// Bits [223:192] are written to bits [191:160] of the return value. \n |
2446 | /// Bits [255:224] are written to bits [255:224] of the return value. |
2447 | /// \returns A 256-bit vector of [8 x float] containing the interleaved values. |
2448 | static __inline __m256 __DEFAULT_FN_ATTRS |
2449 | _mm256_unpackhi_ps(__m256 __a, __m256 __b) |
2450 | { |
2451 | return __builtin_shufflevector((__v8sf)__a, (__v8sf)__b, 2, 10, 2+1, 10+1, 6, 14, 6+1, 14+1); |
2452 | } |
2453 | |
2454 | /// Unpacks the 32-bit vector elements 0, 1, 4 and 5 from each of the |
2455 | /// two 256-bit vectors of [8 x float] and interleaves them into a 256-bit |
2456 | /// vector of [8 x float]. |
2457 | /// |
2458 | /// \headerfile <x86intrin.h> |
2459 | /// |
2460 | /// This intrinsic corresponds to the <c> VUNPCKLPS </c> instruction. |
2461 | /// |
2462 | /// \param __a |
2463 | /// A 256-bit vector of [8 x float]. \n |
2464 | /// Bits [31:0] are written to bits [31:0] of the return value. \n |
2465 | /// Bits [63:32] are written to bits [95:64] of the return value. \n |
2466 | /// Bits [159:128] are written to bits [159:128] of the return value. \n |
2467 | /// Bits [191:160] are written to bits [223:192] of the return value. |
2468 | /// \param __b |
2469 | /// A 256-bit vector of [8 x float]. \n |
2470 | /// Bits [31:0] are written to bits [63:32] of the return value. \n |
2471 | /// Bits [63:32] are written to bits [127:96] of the return value. \n |
2472 | /// Bits [159:128] are written to bits [191:160] of the return value. \n |
2473 | /// Bits [191:160] are written to bits [255:224] of the return value. |
2474 | /// \returns A 256-bit vector of [8 x float] containing the interleaved values. |
2475 | static __inline __m256 __DEFAULT_FN_ATTRS |
2476 | _mm256_unpacklo_ps(__m256 __a, __m256 __b) |
2477 | { |
2478 | return __builtin_shufflevector((__v8sf)__a, (__v8sf)__b, 0, 8, 0+1, 8+1, 4, 12, 4+1, 12+1); |
2479 | } |
2480 | |
2481 | /* Bit Test */ |
2482 | /// Given two 128-bit floating-point vectors of [2 x double], perform an |
2483 | /// element-by-element comparison of the double-precision element in the |
2484 | /// first source vector and the corresponding element in the second source |
2485 | /// vector. |
2486 | /// |
2487 | /// The EFLAGS register is updated as follows: \n |
2488 | /// If there is at least one pair of double-precision elements where the |
2489 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2490 | /// ZF flag is set to 1. \n |
2491 | /// If there is at least one pair of double-precision elements where the |
2492 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2493 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2494 | /// This intrinsic returns the value of the ZF flag. |
2495 | /// |
2496 | /// \headerfile <x86intrin.h> |
2497 | /// |
2498 | /// This intrinsic corresponds to the <c> VTESTPD </c> instruction. |
2499 | /// |
2500 | /// \param __a |
2501 | /// A 128-bit vector of [2 x double]. |
2502 | /// \param __b |
2503 | /// A 128-bit vector of [2 x double]. |
2504 | /// \returns the ZF flag in the EFLAGS register. |
2505 | static __inline int __DEFAULT_FN_ATTRS128 |
2506 | _mm_testz_pd(__m128d __a, __m128d __b) |
2507 | { |
2508 | return __builtin_ia32_vtestzpd((__v2df)__a, (__v2df)__b); |
2509 | } |
2510 | |
2511 | /// Given two 128-bit floating-point vectors of [2 x double], perform an |
2512 | /// element-by-element comparison of the double-precision element in the |
2513 | /// first source vector and the corresponding element in the second source |
2514 | /// vector. |
2515 | /// |
2516 | /// The EFLAGS register is updated as follows: \n |
2517 | /// If there is at least one pair of double-precision elements where the |
2518 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2519 | /// ZF flag is set to 1. \n |
2520 | /// If there is at least one pair of double-precision elements where the |
2521 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2522 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2523 | /// This intrinsic returns the value of the CF flag. |
2524 | /// |
2525 | /// \headerfile <x86intrin.h> |
2526 | /// |
2527 | /// This intrinsic corresponds to the <c> VTESTPD </c> instruction. |
2528 | /// |
2529 | /// \param __a |
2530 | /// A 128-bit vector of [2 x double]. |
2531 | /// \param __b |
2532 | /// A 128-bit vector of [2 x double]. |
2533 | /// \returns the CF flag in the EFLAGS register. |
2534 | static __inline int __DEFAULT_FN_ATTRS128 |
2535 | _mm_testc_pd(__m128d __a, __m128d __b) |
2536 | { |
2537 | return __builtin_ia32_vtestcpd((__v2df)__a, (__v2df)__b); |
2538 | } |
2539 | |
2540 | /// Given two 128-bit floating-point vectors of [2 x double], perform an |
2541 | /// element-by-element comparison of the double-precision element in the |
2542 | /// first source vector and the corresponding element in the second source |
2543 | /// vector. |
2544 | /// |
2545 | /// The EFLAGS register is updated as follows: \n |
2546 | /// If there is at least one pair of double-precision elements where the |
2547 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2548 | /// ZF flag is set to 1. \n |
2549 | /// If there is at least one pair of double-precision elements where the |
2550 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2551 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2552 | /// This intrinsic returns 1 if both the ZF and CF flags are set to 0, |
2553 | /// otherwise it returns 0. |
2554 | /// |
2555 | /// \headerfile <x86intrin.h> |
2556 | /// |
2557 | /// This intrinsic corresponds to the <c> VTESTPD </c> instruction. |
2558 | /// |
2559 | /// \param __a |
2560 | /// A 128-bit vector of [2 x double]. |
2561 | /// \param __b |
2562 | /// A 128-bit vector of [2 x double]. |
2563 | /// \returns 1 if both the ZF and CF flags are set to 0, otherwise returns 0. |
2564 | static __inline int __DEFAULT_FN_ATTRS128 |
2565 | _mm_testnzc_pd(__m128d __a, __m128d __b) |
2566 | { |
2567 | return __builtin_ia32_vtestnzcpd((__v2df)__a, (__v2df)__b); |
2568 | } |
2569 | |
2570 | /// Given two 128-bit floating-point vectors of [4 x float], perform an |
2571 | /// element-by-element comparison of the single-precision element in the |
2572 | /// first source vector and the corresponding element in the second source |
2573 | /// vector. |
2574 | /// |
2575 | /// The EFLAGS register is updated as follows: \n |
2576 | /// If there is at least one pair of single-precision elements where the |
2577 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2578 | /// ZF flag is set to 1. \n |
2579 | /// If there is at least one pair of single-precision elements where the |
2580 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2581 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2582 | /// This intrinsic returns the value of the ZF flag. |
2583 | /// |
2584 | /// \headerfile <x86intrin.h> |
2585 | /// |
2586 | /// This intrinsic corresponds to the <c> VTESTPS </c> instruction. |
2587 | /// |
2588 | /// \param __a |
2589 | /// A 128-bit vector of [4 x float]. |
2590 | /// \param __b |
2591 | /// A 128-bit vector of [4 x float]. |
2592 | /// \returns the ZF flag. |
2593 | static __inline int __DEFAULT_FN_ATTRS128 |
2594 | _mm_testz_ps(__m128 __a, __m128 __b) |
2595 | { |
2596 | return __builtin_ia32_vtestzps((__v4sf)__a, (__v4sf)__b); |
2597 | } |
2598 | |
2599 | /// Given two 128-bit floating-point vectors of [4 x float], perform an |
2600 | /// element-by-element comparison of the single-precision element in the |
2601 | /// first source vector and the corresponding element in the second source |
2602 | /// vector. |
2603 | /// |
2604 | /// The EFLAGS register is updated as follows: \n |
2605 | /// If there is at least one pair of single-precision elements where the |
2606 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2607 | /// ZF flag is set to 1. \n |
2608 | /// If there is at least one pair of single-precision elements where the |
2609 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2610 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2611 | /// This intrinsic returns the value of the CF flag. |
2612 | /// |
2613 | /// \headerfile <x86intrin.h> |
2614 | /// |
2615 | /// This intrinsic corresponds to the <c> VTESTPS </c> instruction. |
2616 | /// |
2617 | /// \param __a |
2618 | /// A 128-bit vector of [4 x float]. |
2619 | /// \param __b |
2620 | /// A 128-bit vector of [4 x float]. |
2621 | /// \returns the CF flag. |
2622 | static __inline int __DEFAULT_FN_ATTRS128 |
2623 | _mm_testc_ps(__m128 __a, __m128 __b) |
2624 | { |
2625 | return __builtin_ia32_vtestcps((__v4sf)__a, (__v4sf)__b); |
2626 | } |
2627 | |
2628 | /// Given two 128-bit floating-point vectors of [4 x float], perform an |
2629 | /// element-by-element comparison of the single-precision element in the |
2630 | /// first source vector and the corresponding element in the second source |
2631 | /// vector. |
2632 | /// |
2633 | /// The EFLAGS register is updated as follows: \n |
2634 | /// If there is at least one pair of single-precision elements where the |
2635 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2636 | /// ZF flag is set to 1. \n |
2637 | /// If there is at least one pair of single-precision elements where the |
2638 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2639 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2640 | /// This intrinsic returns 1 if both the ZF and CF flags are set to 0, |
2641 | /// otherwise it returns 0. |
2642 | /// |
2643 | /// \headerfile <x86intrin.h> |
2644 | /// |
2645 | /// This intrinsic corresponds to the <c> VTESTPS </c> instruction. |
2646 | /// |
2647 | /// \param __a |
2648 | /// A 128-bit vector of [4 x float]. |
2649 | /// \param __b |
2650 | /// A 128-bit vector of [4 x float]. |
2651 | /// \returns 1 if both the ZF and CF flags are set to 0, otherwise returns 0. |
2652 | static __inline int __DEFAULT_FN_ATTRS128 |
2653 | _mm_testnzc_ps(__m128 __a, __m128 __b) |
2654 | { |
2655 | return __builtin_ia32_vtestnzcps((__v4sf)__a, (__v4sf)__b); |
2656 | } |
2657 | |
2658 | /// Given two 256-bit floating-point vectors of [4 x double], perform an |
2659 | /// element-by-element comparison of the double-precision elements in the |
2660 | /// first source vector and the corresponding elements in the second source |
2661 | /// vector. |
2662 | /// |
2663 | /// The EFLAGS register is updated as follows: \n |
2664 | /// If there is at least one pair of double-precision elements where the |
2665 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2666 | /// ZF flag is set to 1. \n |
2667 | /// If there is at least one pair of double-precision elements where the |
2668 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2669 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2670 | /// This intrinsic returns the value of the ZF flag. |
2671 | /// |
2672 | /// \headerfile <x86intrin.h> |
2673 | /// |
2674 | /// This intrinsic corresponds to the <c> VTESTPD </c> instruction. |
2675 | /// |
2676 | /// \param __a |
2677 | /// A 256-bit vector of [4 x double]. |
2678 | /// \param __b |
2679 | /// A 256-bit vector of [4 x double]. |
2680 | /// \returns the ZF flag. |
2681 | static __inline int __DEFAULT_FN_ATTRS |
2682 | _mm256_testz_pd(__m256d __a, __m256d __b) |
2683 | { |
2684 | return __builtin_ia32_vtestzpd256((__v4df)__a, (__v4df)__b); |
2685 | } |
2686 | |
2687 | /// Given two 256-bit floating-point vectors of [4 x double], perform an |
2688 | /// element-by-element comparison of the double-precision elements in the |
2689 | /// first source vector and the corresponding elements in the second source |
2690 | /// vector. |
2691 | /// |
2692 | /// The EFLAGS register is updated as follows: \n |
2693 | /// If there is at least one pair of double-precision elements where the |
2694 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2695 | /// ZF flag is set to 1. \n |
2696 | /// If there is at least one pair of double-precision elements where the |
2697 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2698 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2699 | /// This intrinsic returns the value of the CF flag. |
2700 | /// |
2701 | /// \headerfile <x86intrin.h> |
2702 | /// |
2703 | /// This intrinsic corresponds to the <c> VTESTPD </c> instruction. |
2704 | /// |
2705 | /// \param __a |
2706 | /// A 256-bit vector of [4 x double]. |
2707 | /// \param __b |
2708 | /// A 256-bit vector of [4 x double]. |
2709 | /// \returns the CF flag. |
2710 | static __inline int __DEFAULT_FN_ATTRS |
2711 | _mm256_testc_pd(__m256d __a, __m256d __b) |
2712 | { |
2713 | return __builtin_ia32_vtestcpd256((__v4df)__a, (__v4df)__b); |
2714 | } |
2715 | |
2716 | /// Given two 256-bit floating-point vectors of [4 x double], perform an |
2717 | /// element-by-element comparison of the double-precision elements in the |
2718 | /// first source vector and the corresponding elements in the second source |
2719 | /// vector. |
2720 | /// |
2721 | /// The EFLAGS register is updated as follows: \n |
2722 | /// If there is at least one pair of double-precision elements where the |
2723 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2724 | /// ZF flag is set to 1. \n |
2725 | /// If there is at least one pair of double-precision elements where the |
2726 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2727 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2728 | /// This intrinsic returns 1 if both the ZF and CF flags are set to 0, |
2729 | /// otherwise it returns 0. |
2730 | /// |
2731 | /// \headerfile <x86intrin.h> |
2732 | /// |
2733 | /// This intrinsic corresponds to the <c> VTESTPD </c> instruction. |
2734 | /// |
2735 | /// \param __a |
2736 | /// A 256-bit vector of [4 x double]. |
2737 | /// \param __b |
2738 | /// A 256-bit vector of [4 x double]. |
2739 | /// \returns 1 if both the ZF and CF flags are set to 0, otherwise returns 0. |
2740 | static __inline int __DEFAULT_FN_ATTRS |
2741 | _mm256_testnzc_pd(__m256d __a, __m256d __b) |
2742 | { |
2743 | return __builtin_ia32_vtestnzcpd256((__v4df)__a, (__v4df)__b); |
2744 | } |
2745 | |
2746 | /// Given two 256-bit floating-point vectors of [8 x float], perform an |
2747 | /// element-by-element comparison of the single-precision element in the |
2748 | /// first source vector and the corresponding element in the second source |
2749 | /// vector. |
2750 | /// |
2751 | /// The EFLAGS register is updated as follows: \n |
2752 | /// If there is at least one pair of single-precision elements where the |
2753 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2754 | /// ZF flag is set to 1. \n |
2755 | /// If there is at least one pair of single-precision elements where the |
2756 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2757 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2758 | /// This intrinsic returns the value of the ZF flag. |
2759 | /// |
2760 | /// \headerfile <x86intrin.h> |
2761 | /// |
2762 | /// This intrinsic corresponds to the <c> VTESTPS </c> instruction. |
2763 | /// |
2764 | /// \param __a |
2765 | /// A 256-bit vector of [8 x float]. |
2766 | /// \param __b |
2767 | /// A 256-bit vector of [8 x float]. |
2768 | /// \returns the ZF flag. |
2769 | static __inline int __DEFAULT_FN_ATTRS |
2770 | _mm256_testz_ps(__m256 __a, __m256 __b) |
2771 | { |
2772 | return __builtin_ia32_vtestzps256((__v8sf)__a, (__v8sf)__b); |
2773 | } |
2774 | |
2775 | /// Given two 256-bit floating-point vectors of [8 x float], perform an |
2776 | /// element-by-element comparison of the single-precision element in the |
2777 | /// first source vector and the corresponding element in the second source |
2778 | /// vector. |
2779 | /// |
2780 | /// The EFLAGS register is updated as follows: \n |
2781 | /// If there is at least one pair of single-precision elements where the |
2782 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2783 | /// ZF flag is set to 1. \n |
2784 | /// If there is at least one pair of single-precision elements where the |
2785 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2786 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2787 | /// This intrinsic returns the value of the CF flag. |
2788 | /// |
2789 | /// \headerfile <x86intrin.h> |
2790 | /// |
2791 | /// This intrinsic corresponds to the <c> VTESTPS </c> instruction. |
2792 | /// |
2793 | /// \param __a |
2794 | /// A 256-bit vector of [8 x float]. |
2795 | /// \param __b |
2796 | /// A 256-bit vector of [8 x float]. |
2797 | /// \returns the CF flag. |
2798 | static __inline int __DEFAULT_FN_ATTRS |
2799 | _mm256_testc_ps(__m256 __a, __m256 __b) |
2800 | { |
2801 | return __builtin_ia32_vtestcps256((__v8sf)__a, (__v8sf)__b); |
2802 | } |
2803 | |
2804 | /// Given two 256-bit floating-point vectors of [8 x float], perform an |
2805 | /// element-by-element comparison of the single-precision elements in the |
2806 | /// first source vector and the corresponding elements in the second source |
2807 | /// vector. |
2808 | /// |
2809 | /// The EFLAGS register is updated as follows: \n |
2810 | /// If there is at least one pair of single-precision elements where the |
2811 | /// sign-bits of both elements are 1, the ZF flag is set to 0. Otherwise the |
2812 | /// ZF flag is set to 1. \n |
2813 | /// If there is at least one pair of single-precision elements where the |
2814 | /// sign-bit of the first element is 0 and the sign-bit of the second element |
2815 | /// is 1, the CF flag is set to 0. Otherwise the CF flag is set to 1. \n |
2816 | /// This intrinsic returns 1 if both the ZF and CF flags are set to 0, |
2817 | /// otherwise it returns 0. |
2818 | /// |
2819 | /// \headerfile <x86intrin.h> |
2820 | /// |
2821 | /// This intrinsic corresponds to the <c> VTESTPS </c> instruction. |
2822 | /// |
2823 | /// \param __a |
2824 | /// A 256-bit vector of [8 x float]. |
2825 | /// \param __b |
2826 | /// A 256-bit vector of [8 x float]. |
2827 | /// \returns 1 if both the ZF and CF flags are set to 0, otherwise returns 0. |
2828 | static __inline int __DEFAULT_FN_ATTRS |
2829 | _mm256_testnzc_ps(__m256 __a, __m256 __b) |
2830 | { |
2831 | return __builtin_ia32_vtestnzcps256((__v8sf)__a, (__v8sf)__b); |
2832 | } |
2833 | |
2834 | /// Given two 256-bit integer vectors, perform a bit-by-bit comparison |
2835 | /// of the two source vectors. |
2836 | /// |
2837 | /// The EFLAGS register is updated as follows: \n |
2838 | /// If there is at least one pair of bits where both bits are 1, the ZF flag |
2839 | /// is set to 0. Otherwise the ZF flag is set to 1. \n |
2840 | /// If there is at least one pair of bits where the bit from the first source |
2841 | /// vector is 0 and the bit from the second source vector is 1, the CF flag |
2842 | /// is set to 0. Otherwise the CF flag is set to 1. \n |
2843 | /// This intrinsic returns the value of the ZF flag. |
2844 | /// |
2845 | /// \headerfile <x86intrin.h> |
2846 | /// |
2847 | /// This intrinsic corresponds to the <c> VPTEST </c> instruction. |
2848 | /// |
2849 | /// \param __a |
2850 | /// A 256-bit integer vector. |
2851 | /// \param __b |
2852 | /// A 256-bit integer vector. |
2853 | /// \returns the ZF flag. |
2854 | static __inline int __DEFAULT_FN_ATTRS |
2855 | _mm256_testz_si256(__m256i __a, __m256i __b) |
2856 | { |
2857 | return __builtin_ia32_ptestz256((__v4di)__a, (__v4di)__b); |
2858 | } |
2859 | |
2860 | /// Given two 256-bit integer vectors, perform a bit-by-bit comparison |
2861 | /// of the two source vectors. |
2862 | /// |
2863 | /// The EFLAGS register is updated as follows: \n |
2864 | /// If there is at least one pair of bits where both bits are 1, the ZF flag |
2865 | /// is set to 0. Otherwise the ZF flag is set to 1. \n |
2866 | /// If there is at least one pair of bits where the bit from the first source |
2867 | /// vector is 0 and the bit from the second source vector is 1, the CF flag |
2868 | /// is set to 0. Otherwise the CF flag is set to 1. \n |
2869 | /// This intrinsic returns the value of the CF flag. |
2870 | /// |
2871 | /// \headerfile <x86intrin.h> |
2872 | /// |
2873 | /// This intrinsic corresponds to the <c> VPTEST </c> instruction. |
2874 | /// |
2875 | /// \param __a |
2876 | /// A 256-bit integer vector. |
2877 | /// \param __b |
2878 | /// A 256-bit integer vector. |
2879 | /// \returns the CF flag. |
2880 | static __inline int __DEFAULT_FN_ATTRS |
2881 | _mm256_testc_si256(__m256i __a, __m256i __b) |
2882 | { |
2883 | return __builtin_ia32_ptestc256((__v4di)__a, (__v4di)__b); |
2884 | } |
2885 | |
2886 | /// Given two 256-bit integer vectors, perform a bit-by-bit comparison |
2887 | /// of the two source vectors. |
2888 | /// |
2889 | /// The EFLAGS register is updated as follows: \n |
2890 | /// If there is at least one pair of bits where both bits are 1, the ZF flag |
2891 | /// is set to 0. Otherwise the ZF flag is set to 1. \n |
2892 | /// If there is at least one pair of bits where the bit from the first source |
2893 | /// vector is 0 and the bit from the second source vector is 1, the CF flag |
2894 | /// is set to 0. Otherwise the CF flag is set to 1. \n |
2895 | /// This intrinsic returns 1 if both the ZF and CF flags are set to 0, |
2896 | /// otherwise it returns 0. |
2897 | /// |
2898 | /// \headerfile <x86intrin.h> |
2899 | /// |
2900 | /// This intrinsic corresponds to the <c> VPTEST </c> instruction. |
2901 | /// |
2902 | /// \param __a |
2903 | /// A 256-bit integer vector. |
2904 | /// \param __b |
2905 | /// A 256-bit integer vector. |
2906 | /// \returns 1 if both the ZF and CF flags are set to 0, otherwise returns 0. |
2907 | static __inline int __DEFAULT_FN_ATTRS |
2908 | _mm256_testnzc_si256(__m256i __a, __m256i __b) |
2909 | { |
2910 | return __builtin_ia32_ptestnzc256((__v4di)__a, (__v4di)__b); |
2911 | } |
2912 | |
2913 | /* Vector extract sign mask */ |
2914 | /// Extracts the sign bits of double-precision floating point elements |
2915 | /// in a 256-bit vector of [4 x double] and writes them to the lower order |
2916 | /// bits of the return value. |
2917 | /// |
2918 | /// \headerfile <x86intrin.h> |
2919 | /// |
2920 | /// This intrinsic corresponds to the <c> VMOVMSKPD </c> instruction. |
2921 | /// |
2922 | /// \param __a |
2923 | /// A 256-bit vector of [4 x double] containing the double-precision |
2924 | /// floating point values with sign bits to be extracted. |
2925 | /// \returns The sign bits from the operand, written to bits [3:0]. |
2926 | static __inline int __DEFAULT_FN_ATTRS |
2927 | _mm256_movemask_pd(__m256d __a) |
2928 | { |
2929 | return __builtin_ia32_movmskpd256((__v4df)__a); |
2930 | } |
2931 | |
2932 | /// Extracts the sign bits of single-precision floating point elements |
2933 | /// in a 256-bit vector of [8 x float] and writes them to the lower order |
2934 | /// bits of the return value. |
2935 | /// |
2936 | /// \headerfile <x86intrin.h> |
2937 | /// |
2938 | /// This intrinsic corresponds to the <c> VMOVMSKPS </c> instruction. |
2939 | /// |
2940 | /// \param __a |
2941 | /// A 256-bit vector of [8 x float] containing the single-precision floating |
2942 | /// point values with sign bits to be extracted. |
2943 | /// \returns The sign bits from the operand, written to bits [7:0]. |
2944 | static __inline int __DEFAULT_FN_ATTRS |
2945 | _mm256_movemask_ps(__m256 __a) |
2946 | { |
2947 | return __builtin_ia32_movmskps256((__v8sf)__a); |
2948 | } |
2949 | |
2950 | /* Vector __zero */ |
2951 | /// Zeroes the contents of all XMM or YMM registers. |
2952 | /// |
2953 | /// \headerfile <x86intrin.h> |
2954 | /// |
2955 | /// This intrinsic corresponds to the <c> VZEROALL </c> instruction. |
2956 | static __inline void __attribute__((__always_inline__, __nodebug__, __target__("avx"))) |
2957 | _mm256_zeroall(void) |
2958 | { |
2959 | __builtin_ia32_vzeroall(); |
2960 | } |
2961 | |
2962 | /// Zeroes the upper 128 bits (bits 255:128) of all YMM registers. |
2963 | /// |
2964 | /// \headerfile <x86intrin.h> |
2965 | /// |
2966 | /// This intrinsic corresponds to the <c> VZEROUPPER </c> instruction. |
2967 | static __inline void __attribute__((__always_inline__, __nodebug__, __target__("avx"))) |
2968 | _mm256_zeroupper(void) |
2969 | { |
2970 | __builtin_ia32_vzeroupper(); |
2971 | } |
2972 | |
2973 | /* Vector load with broadcast */ |
2974 | /// Loads a scalar single-precision floating point value from the |
2975 | /// specified address pointed to by \a __a and broadcasts it to the elements |
2976 | /// of a [4 x float] vector. |
2977 | /// |
2978 | /// \headerfile <x86intrin.h> |
2979 | /// |
2980 | /// This intrinsic corresponds to the <c> VBROADCASTSS </c> instruction. |
2981 | /// |
2982 | /// \param __a |
2983 | /// The single-precision floating point value to be broadcast. |
2984 | /// \returns A 128-bit vector of [4 x float] whose 32-bit elements are set |
2985 | /// equal to the broadcast value. |
2986 | static __inline __m128 __DEFAULT_FN_ATTRS128 |
2987 | _mm_broadcast_ss(float const *__a) |
2988 | { |
2989 | float __f = *__a; |
2990 | return __extension__ (__m128)(__v4sf){ __f, __f, __f, __f }; |
2991 | } |
2992 | |
2993 | /// Loads a scalar double-precision floating point value from the |
2994 | /// specified address pointed to by \a __a and broadcasts it to the elements |
2995 | /// of a [4 x double] vector. |
2996 | /// |
2997 | /// \headerfile <x86intrin.h> |
2998 | /// |
2999 | /// This intrinsic corresponds to the <c> VBROADCASTSD </c> instruction. |
3000 | /// |
3001 | /// \param __a |
3002 | /// The double-precision floating point value to be broadcast. |
3003 | /// \returns A 256-bit vector of [4 x double] whose 64-bit elements are set |
3004 | /// equal to the broadcast value. |
3005 | static __inline __m256d __DEFAULT_FN_ATTRS |
3006 | _mm256_broadcast_sd(double const *__a) |
3007 | { |
3008 | double __d = *__a; |
3009 | return __extension__ (__m256d)(__v4df){ __d, __d, __d, __d }; |
3010 | } |
3011 | |
3012 | /// Loads a scalar single-precision floating point value from the |
3013 | /// specified address pointed to by \a __a and broadcasts it to the elements |
3014 | /// of a [8 x float] vector. |
3015 | /// |
3016 | /// \headerfile <x86intrin.h> |
3017 | /// |
3018 | /// This intrinsic corresponds to the <c> VBROADCASTSS </c> instruction. |
3019 | /// |
3020 | /// \param __a |
3021 | /// The single-precision floating point value to be broadcast. |
3022 | /// \returns A 256-bit vector of [8 x float] whose 32-bit elements are set |
3023 | /// equal to the broadcast value. |
3024 | static __inline __m256 __DEFAULT_FN_ATTRS |
3025 | _mm256_broadcast_ss(float const *__a) |
3026 | { |
3027 | float __f = *__a; |
3028 | return __extension__ (__m256)(__v8sf){ __f, __f, __f, __f, __f, __f, __f, __f }; |
3029 | } |
3030 | |
3031 | /// Loads the data from a 128-bit vector of [2 x double] from the |
3032 | /// specified address pointed to by \a __a and broadcasts it to 128-bit |
3033 | /// elements in a 256-bit vector of [4 x double]. |
3034 | /// |
3035 | /// \headerfile <x86intrin.h> |
3036 | /// |
3037 | /// This intrinsic corresponds to the <c> VBROADCASTF128 </c> instruction. |
3038 | /// |
3039 | /// \param __a |
3040 | /// The 128-bit vector of [2 x double] to be broadcast. |
3041 | /// \returns A 256-bit vector of [4 x double] whose 128-bit elements are set |
3042 | /// equal to the broadcast value. |
3043 | static __inline __m256d __DEFAULT_FN_ATTRS |
3044 | _mm256_broadcast_pd(__m128d const *__a) |
3045 | { |
3046 | __m128d __b = _mm_loadu_pd((const double *)__a); |
3047 | return (__m256d)__builtin_shufflevector((__v2df)__b, (__v2df)__b, |
3048 | 0, 1, 0, 1); |
3049 | } |
3050 | |
3051 | /// Loads the data from a 128-bit vector of [4 x float] from the |
3052 | /// specified address pointed to by \a __a and broadcasts it to 128-bit |
3053 | /// elements in a 256-bit vector of [8 x float]. |
3054 | /// |
3055 | /// \headerfile <x86intrin.h> |
3056 | /// |
3057 | /// This intrinsic corresponds to the <c> VBROADCASTF128 </c> instruction. |
3058 | /// |
3059 | /// \param __a |
3060 | /// The 128-bit vector of [4 x float] to be broadcast. |
3061 | /// \returns A 256-bit vector of [8 x float] whose 128-bit elements are set |
3062 | /// equal to the broadcast value. |
3063 | static __inline __m256 __DEFAULT_FN_ATTRS |
3064 | _mm256_broadcast_ps(__m128 const *__a) |
3065 | { |
3066 | __m128 __b = _mm_loadu_ps((const float *)__a); |
3067 | return (__m256)__builtin_shufflevector((__v4sf)__b, (__v4sf)__b, |
3068 | 0, 1, 2, 3, 0, 1, 2, 3); |
3069 | } |
3070 | |
3071 | /* SIMD load ops */ |
3072 | /// Loads 4 double-precision floating point values from a 32-byte aligned |
3073 | /// memory location pointed to by \a __p into a vector of [4 x double]. |
3074 | /// |
3075 | /// \headerfile <x86intrin.h> |
3076 | /// |
3077 | /// This intrinsic corresponds to the <c> VMOVAPD </c> instruction. |
3078 | /// |
3079 | /// \param __p |
3080 | /// A 32-byte aligned pointer to a memory location containing |
3081 | /// double-precision floating point values. |
3082 | /// \returns A 256-bit vector of [4 x double] containing the moved values. |
3083 | static __inline __m256d __DEFAULT_FN_ATTRS |
3084 | _mm256_load_pd(double const *__p) |
3085 | { |
3086 | return *(__m256d *)__p; |
3087 | } |
3088 | |
3089 | /// Loads 8 single-precision floating point values from a 32-byte aligned |
3090 | /// memory location pointed to by \a __p into a vector of [8 x float]. |
3091 | /// |
3092 | /// \headerfile <x86intrin.h> |
3093 | /// |
3094 | /// This intrinsic corresponds to the <c> VMOVAPS </c> instruction. |
3095 | /// |
3096 | /// \param __p |
3097 | /// A 32-byte aligned pointer to a memory location containing float values. |
3098 | /// \returns A 256-bit vector of [8 x float] containing the moved values. |
3099 | static __inline __m256 __DEFAULT_FN_ATTRS |
3100 | _mm256_load_ps(float const *__p) |
3101 | { |
3102 | return *(__m256 *)__p; |
3103 | } |
3104 | |
3105 | /// Loads 4 double-precision floating point values from an unaligned |
3106 | /// memory location pointed to by \a __p into a vector of [4 x double]. |
3107 | /// |
3108 | /// \headerfile <x86intrin.h> |
3109 | /// |
3110 | /// This intrinsic corresponds to the <c> VMOVUPD </c> instruction. |
3111 | /// |
3112 | /// \param __p |
3113 | /// A pointer to a memory location containing double-precision floating |
3114 | /// point values. |
3115 | /// \returns A 256-bit vector of [4 x double] containing the moved values. |
3116 | static __inline __m256d __DEFAULT_FN_ATTRS |
3117 | _mm256_loadu_pd(double const *__p) |
3118 | { |
3119 | struct __loadu_pd { |
3120 | __m256d_u __v; |
3121 | } __attribute__((__packed__, __may_alias__)); |
3122 | return ((struct __loadu_pd*)__p)->__v; |
3123 | } |
3124 | |
3125 | /// Loads 8 single-precision floating point values from an unaligned |
3126 | /// memory location pointed to by \a __p into a vector of [8 x float]. |
3127 | /// |
3128 | /// \headerfile <x86intrin.h> |
3129 | /// |
3130 | /// This intrinsic corresponds to the <c> VMOVUPS </c> instruction. |
3131 | /// |
3132 | /// \param __p |
3133 | /// A pointer to a memory location containing single-precision floating |
3134 | /// point values. |
3135 | /// \returns A 256-bit vector of [8 x float] containing the moved values. |
3136 | static __inline __m256 __DEFAULT_FN_ATTRS |
3137 | _mm256_loadu_ps(float const *__p) |
3138 | { |
3139 | struct __loadu_ps { |
3140 | __m256_u __v; |
3141 | } __attribute__((__packed__, __may_alias__)); |
3142 | return ((struct __loadu_ps*)__p)->__v; |
3143 | } |
3144 | |
3145 | /// Loads 256 bits of integer data from a 32-byte aligned memory |
3146 | /// location pointed to by \a __p into elements of a 256-bit integer vector. |
3147 | /// |
3148 | /// \headerfile <x86intrin.h> |
3149 | /// |
3150 | /// This intrinsic corresponds to the <c> VMOVDQA </c> instruction. |
3151 | /// |
3152 | /// \param __p |
3153 | /// A 32-byte aligned pointer to a 256-bit integer vector containing integer |
3154 | /// values. |
3155 | /// \returns A 256-bit integer vector containing the moved values. |
3156 | static __inline __m256i __DEFAULT_FN_ATTRS |
3157 | _mm256_load_si256(__m256i const *__p) |
3158 | { |
3159 | return *__p; |
3160 | } |
3161 | |
3162 | /// Loads 256 bits of integer data from an unaligned memory location |
3163 | /// pointed to by \a __p into a 256-bit integer vector. |
3164 | /// |
3165 | /// \headerfile <x86intrin.h> |
3166 | /// |
3167 | /// This intrinsic corresponds to the <c> VMOVDQU </c> instruction. |
3168 | /// |
3169 | /// \param __p |
3170 | /// A pointer to a 256-bit integer vector containing integer values. |
3171 | /// \returns A 256-bit integer vector containing the moved values. |
3172 | static __inline __m256i __DEFAULT_FN_ATTRS |
3173 | _mm256_loadu_si256(__m256i_u const *__p) |
3174 | { |
3175 | struct __loadu_si256 { |
3176 | __m256i_u __v; |
3177 | } __attribute__((__packed__, __may_alias__)); |
3178 | return ((struct __loadu_si256*)__p)->__v; |
3179 | } |
3180 | |
3181 | /// Loads 256 bits of integer data from an unaligned memory location |
3182 | /// pointed to by \a __p into a 256-bit integer vector. This intrinsic may |
3183 | /// perform better than \c _mm256_loadu_si256 when the data crosses a cache |
3184 | /// line boundary. |
3185 | /// |
3186 | /// \headerfile <x86intrin.h> |
3187 | /// |
3188 | /// This intrinsic corresponds to the <c> VLDDQU </c> instruction. |
3189 | /// |
3190 | /// \param __p |
3191 | /// A pointer to a 256-bit integer vector containing integer values. |
3192 | /// \returns A 256-bit integer vector containing the moved values. |
3193 | static __inline __m256i __DEFAULT_FN_ATTRS |
3194 | _mm256_lddqu_si256(__m256i const *__p) |
3195 | { |
3196 | return (__m256i)__builtin_ia32_lddqu256((char const *)__p); |
3197 | } |
3198 | |
3199 | /* SIMD store ops */ |
3200 | /// Stores double-precision floating point values from a 256-bit vector |
3201 | /// of [4 x double] to a 32-byte aligned memory location pointed to by |
3202 | /// \a __p. |
3203 | /// |
3204 | /// \headerfile <x86intrin.h> |
3205 | /// |
3206 | /// This intrinsic corresponds to the <c> VMOVAPD </c> instruction. |
3207 | /// |
3208 | /// \param __p |
3209 | /// A 32-byte aligned pointer to a memory location that will receive the |
3210 | /// double-precision floaing point values. |
3211 | /// \param __a |
3212 | /// A 256-bit vector of [4 x double] containing the values to be moved. |
3213 | static __inline void __DEFAULT_FN_ATTRS |
3214 | _mm256_store_pd(double *__p, __m256d __a) |
3215 | { |
3216 | *(__m256d *)__p = __a; |
3217 | } |
3218 | |
3219 | /// Stores single-precision floating point values from a 256-bit vector |
3220 | /// of [8 x float] to a 32-byte aligned memory location pointed to by \a __p. |
3221 | /// |
3222 | /// \headerfile <x86intrin.h> |
3223 | /// |
3224 | /// This intrinsic corresponds to the <c> VMOVAPS </c> instruction. |
3225 | /// |
3226 | /// \param __p |
3227 | /// A 32-byte aligned pointer to a memory location that will receive the |
3228 | /// float values. |
3229 | /// \param __a |
3230 | /// A 256-bit vector of [8 x float] containing the values to be moved. |
3231 | static __inline void __DEFAULT_FN_ATTRS |
3232 | _mm256_store_ps(float *__p, __m256 __a) |
3233 | { |
3234 | *(__m256 *)__p = __a; |
3235 | } |
3236 | |
3237 | /// Stores double-precision floating point values from a 256-bit vector |
3238 | /// of [4 x double] to an unaligned memory location pointed to by \a __p. |
3239 | /// |
3240 | /// \headerfile <x86intrin.h> |
3241 | /// |
3242 | /// This intrinsic corresponds to the <c> VMOVUPD </c> instruction. |
3243 | /// |
3244 | /// \param __p |
3245 | /// A pointer to a memory location that will receive the double-precision |
3246 | /// floating point values. |
3247 | /// \param __a |
3248 | /// A 256-bit vector of [4 x double] containing the values to be moved. |
3249 | static __inline void __DEFAULT_FN_ATTRS |
3250 | _mm256_storeu_pd(double *__p, __m256d __a) |
3251 | { |
3252 | struct __storeu_pd { |
3253 | __m256d_u __v; |
3254 | } __attribute__((__packed__, __may_alias__)); |
3255 | ((struct __storeu_pd*)__p)->__v = __a; |
3256 | } |
3257 | |
3258 | /// Stores single-precision floating point values from a 256-bit vector |
3259 | /// of [8 x float] to an unaligned memory location pointed to by \a __p. |
3260 | /// |
3261 | /// \headerfile <x86intrin.h> |
3262 | /// |
3263 | /// This intrinsic corresponds to the <c> VMOVUPS </c> instruction. |
3264 | /// |
3265 | /// \param __p |
3266 | /// A pointer to a memory location that will receive the float values. |
3267 | /// \param __a |
3268 | /// A 256-bit vector of [8 x float] containing the values to be moved. |
3269 | static __inline void __DEFAULT_FN_ATTRS |
3270 | _mm256_storeu_ps(float *__p, __m256 __a) |
3271 | { |
3272 | struct __storeu_ps { |
3273 | __m256_u __v; |
3274 | } __attribute__((__packed__, __may_alias__)); |
3275 | ((struct __storeu_ps*)__p)->__v = __a; |
3276 | } |
3277 | |
3278 | /// Stores integer values from a 256-bit integer vector to a 32-byte |
3279 | /// aligned memory location pointed to by \a __p. |
3280 | /// |
3281 | /// \headerfile <x86intrin.h> |
3282 | /// |
3283 | /// This intrinsic corresponds to the <c> VMOVDQA </c> instruction. |
3284 | /// |
3285 | /// \param __p |
3286 | /// A 32-byte aligned pointer to a memory location that will receive the |
3287 | /// integer values. |
3288 | /// \param __a |
3289 | /// A 256-bit integer vector containing the values to be moved. |
3290 | static __inline void __DEFAULT_FN_ATTRS |
3291 | _mm256_store_si256(__m256i *__p, __m256i __a) |
3292 | { |
3293 | *__p = __a; |
3294 | } |
3295 | |
3296 | /// Stores integer values from a 256-bit integer vector to an unaligned |
3297 | /// memory location pointed to by \a __p. |
3298 | /// |
3299 | /// \headerfile <x86intrin.h> |
3300 | /// |
3301 | /// This intrinsic corresponds to the <c> VMOVDQU </c> instruction. |
3302 | /// |
3303 | /// \param __p |
3304 | /// A pointer to a memory location that will receive the integer values. |
3305 | /// \param __a |
3306 | /// A 256-bit integer vector containing the values to be moved. |
3307 | static __inline void __DEFAULT_FN_ATTRS |
3308 | _mm256_storeu_si256(__m256i_u *__p, __m256i __a) |
3309 | { |
3310 | struct __storeu_si256 { |
3311 | __m256i_u __v; |
3312 | } __attribute__((__packed__, __may_alias__)); |
3313 | ((struct __storeu_si256*)__p)->__v = __a; |
3314 | } |
3315 | |
3316 | /* Conditional load ops */ |
3317 | /// Conditionally loads double-precision floating point elements from a |
3318 | /// memory location pointed to by \a __p into a 128-bit vector of |
3319 | /// [2 x double], depending on the mask bits associated with each data |
3320 | /// element. |
3321 | /// |
3322 | /// \headerfile <x86intrin.h> |
3323 | /// |
3324 | /// This intrinsic corresponds to the <c> VMASKMOVPD </c> instruction. |
3325 | /// |
3326 | /// \param __p |
3327 | /// A pointer to a memory location that contains the double-precision |
3328 | /// floating point values. |
3329 | /// \param __m |
3330 | /// A 128-bit integer vector containing the mask. The most significant bit of |
3331 | /// each data element represents the mask bits. If a mask bit is zero, the |
3332 | /// corresponding value in the memory location is not loaded and the |
3333 | /// corresponding field in the return value is set to zero. |
3334 | /// \returns A 128-bit vector of [2 x double] containing the loaded values. |
3335 | static __inline __m128d __DEFAULT_FN_ATTRS128 |
3336 | _mm_maskload_pd(double const *__p, __m128i __m) |
3337 | { |
3338 | return (__m128d)__builtin_ia32_maskloadpd((const __v2df *)__p, (__v2di)__m); |
3339 | } |
3340 | |
3341 | /// Conditionally loads double-precision floating point elements from a |
3342 | /// memory location pointed to by \a __p into a 256-bit vector of |
3343 | /// [4 x double], depending on the mask bits associated with each data |
3344 | /// element. |
3345 | /// |
3346 | /// \headerfile <x86intrin.h> |
3347 | /// |
3348 | /// This intrinsic corresponds to the <c> VMASKMOVPD </c> instruction. |
3349 | /// |
3350 | /// \param __p |
3351 | /// A pointer to a memory location that contains the double-precision |
3352 | /// floating point values. |
3353 | /// \param __m |
3354 | /// A 256-bit integer vector of [4 x quadword] containing the mask. The most |
3355 | /// significant bit of each quadword element represents the mask bits. If a |
3356 | /// mask bit is zero, the corresponding value in the memory location is not |
3357 | /// loaded and the corresponding field in the return value is set to zero. |
3358 | /// \returns A 256-bit vector of [4 x double] containing the loaded values. |
3359 | static __inline __m256d __DEFAULT_FN_ATTRS |
3360 | _mm256_maskload_pd(double const *__p, __m256i __m) |
3361 | { |
3362 | return (__m256d)__builtin_ia32_maskloadpd256((const __v4df *)__p, |
3363 | (__v4di)__m); |
3364 | } |
3365 | |
3366 | /// Conditionally loads single-precision floating point elements from a |
3367 | /// memory location pointed to by \a __p into a 128-bit vector of |
3368 | /// [4 x float], depending on the mask bits associated with each data |
3369 | /// element. |
3370 | /// |
3371 | /// \headerfile <x86intrin.h> |
3372 | /// |
3373 | /// This intrinsic corresponds to the <c> VMASKMOVPS </c> instruction. |
3374 | /// |
3375 | /// \param __p |
3376 | /// A pointer to a memory location that contains the single-precision |
3377 | /// floating point values. |
3378 | /// \param __m |
3379 | /// A 128-bit integer vector containing the mask. The most significant bit of |
3380 | /// each data element represents the mask bits. If a mask bit is zero, the |
3381 | /// corresponding value in the memory location is not loaded and the |
3382 | /// corresponding field in the return value is set to zero. |
3383 | /// \returns A 128-bit vector of [4 x float] containing the loaded values. |
3384 | static __inline __m128 __DEFAULT_FN_ATTRS128 |
3385 | _mm_maskload_ps(float const *__p, __m128i __m) |
3386 | { |
3387 | return (__m128)__builtin_ia32_maskloadps((const __v4sf *)__p, (__v4si)__m); |
3388 | } |
3389 | |
3390 | /// Conditionally loads single-precision floating point elements from a |
3391 | /// memory location pointed to by \a __p into a 256-bit vector of |
3392 | /// [8 x float], depending on the mask bits associated with each data |
3393 | /// element. |
3394 | /// |
3395 | /// \headerfile <x86intrin.h> |
3396 | /// |
3397 | /// This intrinsic corresponds to the <c> VMASKMOVPS </c> instruction. |
3398 | /// |
3399 | /// \param __p |
3400 | /// A pointer to a memory location that contains the single-precision |
3401 | /// floating point values. |
3402 | /// \param __m |
3403 | /// A 256-bit integer vector of [8 x dword] containing the mask. The most |
3404 | /// significant bit of each dword element represents the mask bits. If a mask |
3405 | /// bit is zero, the corresponding value in the memory location is not loaded |
3406 | /// and the corresponding field in the return value is set to zero. |
3407 | /// \returns A 256-bit vector of [8 x float] containing the loaded values. |
3408 | static __inline __m256 __DEFAULT_FN_ATTRS |
3409 | _mm256_maskload_ps(float const *__p, __m256i __m) |
3410 | { |
3411 | return (__m256)__builtin_ia32_maskloadps256((const __v8sf *)__p, (__v8si)__m); |
3412 | } |
3413 | |
3414 | /* Conditional store ops */ |
3415 | /// Moves single-precision floating point values from a 256-bit vector |
3416 | /// of [8 x float] to a memory location pointed to by \a __p, according to |
3417 | /// the specified mask. |
3418 | /// |
3419 | /// \headerfile <x86intrin.h> |
3420 | /// |
3421 | /// This intrinsic corresponds to the <c> VMASKMOVPS </c> instruction. |
3422 | /// |
3423 | /// \param __p |
3424 | /// A pointer to a memory location that will receive the float values. |
3425 | /// \param __m |
3426 | /// A 256-bit integer vector of [8 x dword] containing the mask. The most |
3427 | /// significant bit of each dword element in the mask vector represents the |
3428 | /// mask bits. If a mask bit is zero, the corresponding value from vector |
3429 | /// \a __a is not stored and the corresponding field in the memory location |
3430 | /// pointed to by \a __p is not changed. |
3431 | /// \param __a |
3432 | /// A 256-bit vector of [8 x float] containing the values to be stored. |
3433 | static __inline void __DEFAULT_FN_ATTRS |
3434 | _mm256_maskstore_ps(float *__p, __m256i __m, __m256 __a) |
3435 | { |
3436 | __builtin_ia32_maskstoreps256((__v8sf *)__p, (__v8si)__m, (__v8sf)__a); |
3437 | } |
3438 | |
3439 | /// Moves double-precision values from a 128-bit vector of [2 x double] |
3440 | /// to a memory location pointed to by \a __p, according to the specified |
3441 | /// mask. |
3442 | /// |
3443 | /// \headerfile <x86intrin.h> |
3444 | /// |
3445 | /// This intrinsic corresponds to the <c> VMASKMOVPD </c> instruction. |
3446 | /// |
3447 | /// \param __p |
3448 | /// A pointer to a memory location that will receive the float values. |
3449 | /// \param __m |
3450 | /// A 128-bit integer vector containing the mask. The most significant bit of |
3451 | /// each field in the mask vector represents the mask bits. If a mask bit is |
3452 | /// zero, the corresponding value from vector \a __a is not stored and the |
3453 | /// corresponding field in the memory location pointed to by \a __p is not |
3454 | /// changed. |
3455 | /// \param __a |
3456 | /// A 128-bit vector of [2 x double] containing the values to be stored. |
3457 | static __inline void __DEFAULT_FN_ATTRS128 |
3458 | _mm_maskstore_pd(double *__p, __m128i __m, __m128d __a) |
3459 | { |
3460 | __builtin_ia32_maskstorepd((__v2df *)__p, (__v2di)__m, (__v2df)__a); |
3461 | } |
3462 | |
3463 | /// Moves double-precision values from a 256-bit vector of [4 x double] |
3464 | /// to a memory location pointed to by \a __p, according to the specified |
3465 | /// mask. |
3466 | /// |
3467 | /// \headerfile <x86intrin.h> |
3468 | /// |
3469 | /// This intrinsic corresponds to the <c> VMASKMOVPD </c> instruction. |
3470 | /// |
3471 | /// \param __p |
3472 | /// A pointer to a memory location that will receive the float values. |
3473 | /// \param __m |
3474 | /// A 256-bit integer vector of [4 x quadword] containing the mask. The most |
3475 | /// significant bit of each quadword element in the mask vector represents |
3476 | /// the mask bits. If a mask bit is zero, the corresponding value from vector |
3477 | /// __a is not stored and the corresponding field in the memory location |
3478 | /// pointed to by \a __p is not changed. |
3479 | /// \param __a |
3480 | /// A 256-bit vector of [4 x double] containing the values to be stored. |
3481 | static __inline void __DEFAULT_FN_ATTRS |
3482 | _mm256_maskstore_pd(double *__p, __m256i __m, __m256d __a) |
3483 | { |
3484 | __builtin_ia32_maskstorepd256((__v4df *)__p, (__v4di)__m, (__v4df)__a); |
3485 | } |
3486 | |
3487 | /// Moves single-precision floating point values from a 128-bit vector |
3488 | /// of [4 x float] to a memory location pointed to by \a __p, according to |
3489 | /// the specified mask. |
3490 | /// |
3491 | /// \headerfile <x86intrin.h> |
3492 | /// |
3493 | /// This intrinsic corresponds to the <c> VMASKMOVPS </c> instruction. |
3494 | /// |
3495 | /// \param __p |
3496 | /// A pointer to a memory location that will receive the float values. |
3497 | /// \param __m |
3498 | /// A 128-bit integer vector containing the mask. The most significant bit of |
3499 | /// each field in the mask vector represents the mask bits. If a mask bit is |
3500 | /// zero, the corresponding value from vector __a is not stored and the |
3501 | /// corresponding field in the memory location pointed to by \a __p is not |
3502 | /// changed. |
3503 | /// \param __a |
3504 | /// A 128-bit vector of [4 x float] containing the values to be stored. |
3505 | static __inline void __DEFAULT_FN_ATTRS128 |
3506 | _mm_maskstore_ps(float *__p, __m128i __m, __m128 __a) |
3507 | { |
3508 | __builtin_ia32_maskstoreps((__v4sf *)__p, (__v4si)__m, (__v4sf)__a); |
3509 | } |
3510 | |
3511 | /* Cacheability support ops */ |
3512 | /// Moves integer data from a 256-bit integer vector to a 32-byte |
3513 | /// aligned memory location. To minimize caching, the data is flagged as |
3514 | /// non-temporal (unlikely to be used again soon). |
3515 | /// |
3516 | /// \headerfile <x86intrin.h> |
3517 | /// |
3518 | /// This intrinsic corresponds to the <c> VMOVNTDQ </c> instruction. |
3519 | /// |
3520 | /// \param __a |
3521 | /// A pointer to a 32-byte aligned memory location that will receive the |
3522 | /// integer values. |
3523 | /// \param __b |
3524 | /// A 256-bit integer vector containing the values to be moved. |
3525 | static __inline void __DEFAULT_FN_ATTRS |
3526 | _mm256_stream_si256(__m256i *__a, __m256i __b) |
3527 | { |
3528 | typedef __v4di __v4di_aligned __attribute__((aligned(32))); |
3529 | __builtin_nontemporal_store((__v4di_aligned)__b, (__v4di_aligned*)__a); |
3530 | } |
3531 | |
3532 | /// Moves double-precision values from a 256-bit vector of [4 x double] |
3533 | /// to a 32-byte aligned memory location. To minimize caching, the data is |
3534 | /// flagged as non-temporal (unlikely to be used again soon). |
3535 | /// |
3536 | /// \headerfile <x86intrin.h> |
3537 | /// |
3538 | /// This intrinsic corresponds to the <c> VMOVNTPD </c> instruction. |
3539 | /// |
3540 | /// \param __a |
3541 | /// A pointer to a 32-byte aligned memory location that will receive the |
3542 | /// double-precision floating-point values. |
3543 | /// \param __b |
3544 | /// A 256-bit vector of [4 x double] containing the values to be moved. |
3545 | static __inline void __DEFAULT_FN_ATTRS |
3546 | _mm256_stream_pd(double *__a, __m256d __b) |
3547 | { |
3548 | typedef __v4df __v4df_aligned __attribute__((aligned(32))); |
3549 | __builtin_nontemporal_store((__v4df_aligned)__b, (__v4df_aligned*)__a); |
3550 | } |
3551 | |
3552 | /// Moves single-precision floating point values from a 256-bit vector |
3553 | /// of [8 x float] to a 32-byte aligned memory location. To minimize |
3554 | /// caching, the data is flagged as non-temporal (unlikely to be used again |
3555 | /// soon). |
3556 | /// |
3557 | /// \headerfile <x86intrin.h> |
3558 | /// |
3559 | /// This intrinsic corresponds to the <c> VMOVNTPS </c> instruction. |
3560 | /// |
3561 | /// \param __p |
3562 | /// A pointer to a 32-byte aligned memory location that will receive the |
3563 | /// single-precision floating point values. |
3564 | /// \param __a |
3565 | /// A 256-bit vector of [8 x float] containing the values to be moved. |
3566 | static __inline void __DEFAULT_FN_ATTRS |
3567 | _mm256_stream_ps(float *__p, __m256 __a) |
3568 | { |
3569 | typedef __v8sf __v8sf_aligned __attribute__((aligned(32))); |
3570 | __builtin_nontemporal_store((__v8sf_aligned)__a, (__v8sf_aligned*)__p); |
3571 | } |
3572 | |
3573 | /* Create vectors */ |
3574 | /// Create a 256-bit vector of [4 x double] with undefined values. |
3575 | /// |
3576 | /// \headerfile <x86intrin.h> |
3577 | /// |
3578 | /// This intrinsic has no corresponding instruction. |
3579 | /// |
3580 | /// \returns A 256-bit vector of [4 x double] containing undefined values. |
3581 | static __inline__ __m256d __DEFAULT_FN_ATTRS |
3582 | _mm256_undefined_pd(void) |
3583 | { |
3584 | return (__m256d)__builtin_ia32_undef256(); |
3585 | } |
3586 | |
3587 | /// Create a 256-bit vector of [8 x float] with undefined values. |
3588 | /// |
3589 | /// \headerfile <x86intrin.h> |
3590 | /// |
3591 | /// This intrinsic has no corresponding instruction. |
3592 | /// |
3593 | /// \returns A 256-bit vector of [8 x float] containing undefined values. |
3594 | static __inline__ __m256 __DEFAULT_FN_ATTRS |
3595 | _mm256_undefined_ps(void) |
3596 | { |
3597 | return (__m256)__builtin_ia32_undef256(); |
3598 | } |
3599 | |
3600 | /// Create a 256-bit integer vector with undefined values. |
3601 | /// |
3602 | /// \headerfile <x86intrin.h> |
3603 | /// |
3604 | /// This intrinsic has no corresponding instruction. |
3605 | /// |
3606 | /// \returns A 256-bit integer vector containing undefined values. |
3607 | static __inline__ __m256i __DEFAULT_FN_ATTRS |
3608 | _mm256_undefined_si256(void) |
3609 | { |
3610 | return (__m256i)__builtin_ia32_undef256(); |
3611 | } |
3612 | |
3613 | /// Constructs a 256-bit floating-point vector of [4 x double] |
3614 | /// initialized with the specified double-precision floating-point values. |
3615 | /// |
3616 | /// \headerfile <x86intrin.h> |
3617 | /// |
3618 | /// This intrinsic corresponds to the <c> VUNPCKLPD+VINSERTF128 </c> |
3619 | /// instruction. |
3620 | /// |
3621 | /// \param __a |
3622 | /// A double-precision floating-point value used to initialize bits [255:192] |
3623 | /// of the result. |
3624 | /// \param __b |
3625 | /// A double-precision floating-point value used to initialize bits [191:128] |
3626 | /// of the result. |
3627 | /// \param __c |
3628 | /// A double-precision floating-point value used to initialize bits [127:64] |
3629 | /// of the result. |
3630 | /// \param __d |
3631 | /// A double-precision floating-point value used to initialize bits [63:0] |
3632 | /// of the result. |
3633 | /// \returns An initialized 256-bit floating-point vector of [4 x double]. |
3634 | static __inline __m256d __DEFAULT_FN_ATTRS |
3635 | _mm256_set_pd(double __a, double __b, double __c, double __d) |
3636 | { |
3637 | return __extension__ (__m256d){ __d, __c, __b, __a }; |
3638 | } |
3639 | |
3640 | /// Constructs a 256-bit floating-point vector of [8 x float] initialized |
3641 | /// with the specified single-precision floating-point values. |
3642 | /// |
3643 | /// \headerfile <x86intrin.h> |
3644 | /// |
3645 | /// This intrinsic is a utility function and does not correspond to a specific |
3646 | /// instruction. |
3647 | /// |
3648 | /// \param __a |
3649 | /// A single-precision floating-point value used to initialize bits [255:224] |
3650 | /// of the result. |
3651 | /// \param __b |
3652 | /// A single-precision floating-point value used to initialize bits [223:192] |
3653 | /// of the result. |
3654 | /// \param __c |
3655 | /// A single-precision floating-point value used to initialize bits [191:160] |
3656 | /// of the result. |
3657 | /// \param __d |
3658 | /// A single-precision floating-point value used to initialize bits [159:128] |
3659 | /// of the result. |
3660 | /// \param __e |
3661 | /// A single-precision floating-point value used to initialize bits [127:96] |
3662 | /// of the result. |
3663 | /// \param __f |
3664 | /// A single-precision floating-point value used to initialize bits [95:64] |
3665 | /// of the result. |
3666 | /// \param __g |
3667 | /// A single-precision floating-point value used to initialize bits [63:32] |
3668 | /// of the result. |
3669 | /// \param __h |
3670 | /// A single-precision floating-point value used to initialize bits [31:0] |
3671 | /// of the result. |
3672 | /// \returns An initialized 256-bit floating-point vector of [8 x float]. |
3673 | static __inline __m256 __DEFAULT_FN_ATTRS |
3674 | _mm256_set_ps(float __a, float __b, float __c, float __d, |
3675 | float __e, float __f, float __g, float __h) |
3676 | { |
3677 | return __extension__ (__m256){ __h, __g, __f, __e, __d, __c, __b, __a }; |
3678 | } |
3679 | |
3680 | /// Constructs a 256-bit integer vector initialized with the specified |
3681 | /// 32-bit integral values. |
3682 | /// |
3683 | /// \headerfile <x86intrin.h> |
3684 | /// |
3685 | /// This intrinsic is a utility function and does not correspond to a specific |
3686 | /// instruction. |
3687 | /// |
3688 | /// \param __i0 |
3689 | /// A 32-bit integral value used to initialize bits [255:224] of the result. |
3690 | /// \param __i1 |
3691 | /// A 32-bit integral value used to initialize bits [223:192] of the result. |
3692 | /// \param __i2 |
3693 | /// A 32-bit integral value used to initialize bits [191:160] of the result. |
3694 | /// \param __i3 |
3695 | /// A 32-bit integral value used to initialize bits [159:128] of the result. |
3696 | /// \param __i4 |
3697 | /// A 32-bit integral value used to initialize bits [127:96] of the result. |
3698 | /// \param __i5 |
3699 | /// A 32-bit integral value used to initialize bits [95:64] of the result. |
3700 | /// \param __i6 |
3701 | /// A 32-bit integral value used to initialize bits [63:32] of the result. |
3702 | /// \param __i7 |
3703 | /// A 32-bit integral value used to initialize bits [31:0] of the result. |
3704 | /// \returns An initialized 256-bit integer vector. |
3705 | static __inline __m256i __DEFAULT_FN_ATTRS |
3706 | _mm256_set_epi32(int __i0, int __i1, int __i2, int __i3, |
3707 | int __i4, int __i5, int __i6, int __i7) |
3708 | { |
3709 | return __extension__ (__m256i)(__v8si){ __i7, __i6, __i5, __i4, __i3, __i2, __i1, __i0 }; |
3710 | } |
3711 | |
3712 | /// Constructs a 256-bit integer vector initialized with the specified |
3713 | /// 16-bit integral values. |
3714 | /// |
3715 | /// \headerfile <x86intrin.h> |
3716 | /// |
3717 | /// This intrinsic is a utility function and does not correspond to a specific |
3718 | /// instruction. |
3719 | /// |
3720 | /// \param __w15 |
3721 | /// A 16-bit integral value used to initialize bits [255:240] of the result. |
3722 | /// \param __w14 |
3723 | /// A 16-bit integral value used to initialize bits [239:224] of the result. |
3724 | /// \param __w13 |
3725 | /// A 16-bit integral value used to initialize bits [223:208] of the result. |
3726 | /// \param __w12 |
3727 | /// A 16-bit integral value used to initialize bits [207:192] of the result. |
3728 | /// \param __w11 |
3729 | /// A 16-bit integral value used to initialize bits [191:176] of the result. |
3730 | /// \param __w10 |
3731 | /// A 16-bit integral value used to initialize bits [175:160] of the result. |
3732 | /// \param __w09 |
3733 | /// A 16-bit integral value used to initialize bits [159:144] of the result. |
3734 | /// \param __w08 |
3735 | /// A 16-bit integral value used to initialize bits [143:128] of the result. |
3736 | /// \param __w07 |
3737 | /// A 16-bit integral value used to initialize bits [127:112] of the result. |
3738 | /// \param __w06 |
3739 | /// A 16-bit integral value used to initialize bits [111:96] of the result. |
3740 | /// \param __w05 |
3741 | /// A 16-bit integral value used to initialize bits [95:80] of the result. |
3742 | /// \param __w04 |
3743 | /// A 16-bit integral value used to initialize bits [79:64] of the result. |
3744 | /// \param __w03 |
3745 | /// A 16-bit integral value used to initialize bits [63:48] of the result. |
3746 | /// \param __w02 |
3747 | /// A 16-bit integral value used to initialize bits [47:32] of the result. |
3748 | /// \param __w01 |
3749 | /// A 16-bit integral value used to initialize bits [31:16] of the result. |
3750 | /// \param __w00 |
3751 | /// A 16-bit integral value used to initialize bits [15:0] of the result. |
3752 | /// \returns An initialized 256-bit integer vector. |
3753 | static __inline __m256i __DEFAULT_FN_ATTRS |
3754 | _mm256_set_epi16(short __w15, short __w14, short __w13, short __w12, |
3755 | short __w11, short __w10, short __w09, short __w08, |
3756 | short __w07, short __w06, short __w05, short __w04, |
3757 | short __w03, short __w02, short __w01, short __w00) |
3758 | { |
3759 | return __extension__ (__m256i)(__v16hi){ __w00, __w01, __w02, __w03, __w04, __w05, __w06, |
3760 | __w07, __w08, __w09, __w10, __w11, __w12, __w13, __w14, __w15 }; |
3761 | } |
3762 | |
3763 | /// Constructs a 256-bit integer vector initialized with the specified |
3764 | /// 8-bit integral values. |
3765 | /// |
3766 | /// \headerfile <x86intrin.h> |
3767 | /// |
3768 | /// This intrinsic is a utility function and does not correspond to a specific |
3769 | /// instruction. |
3770 | /// |
3771 | /// \param __b31 |
3772 | /// An 8-bit integral value used to initialize bits [255:248] of the result. |
3773 | /// \param __b30 |
3774 | /// An 8-bit integral value used to initialize bits [247:240] of the result. |
3775 | /// \param __b29 |
3776 | /// An 8-bit integral value used to initialize bits [239:232] of the result. |
3777 | /// \param __b28 |
3778 | /// An 8-bit integral value used to initialize bits [231:224] of the result. |
3779 | /// \param __b27 |
3780 | /// An 8-bit integral value used to initialize bits [223:216] of the result. |
3781 | /// \param __b26 |
3782 | /// An 8-bit integral value used to initialize bits [215:208] of the result. |
3783 | /// \param __b25 |
3784 | /// An 8-bit integral value used to initialize bits [207:200] of the result. |
3785 | /// \param __b24 |
3786 | /// An 8-bit integral value used to initialize bits [199:192] of the result. |
3787 | /// \param __b23 |
3788 | /// An 8-bit integral value used to initialize bits [191:184] of the result. |
3789 | /// \param __b22 |
3790 | /// An 8-bit integral value used to initialize bits [183:176] of the result. |
3791 | /// \param __b21 |
3792 | /// An 8-bit integral value used to initialize bits [175:168] of the result. |
3793 | /// \param __b20 |
3794 | /// An 8-bit integral value used to initialize bits [167:160] of the result. |
3795 | /// \param __b19 |
3796 | /// An 8-bit integral value used to initialize bits [159:152] of the result. |
3797 | /// \param __b18 |
3798 | /// An 8-bit integral value used to initialize bits [151:144] of the result. |
3799 | /// \param __b17 |
3800 | /// An 8-bit integral value used to initialize bits [143:136] of the result. |
3801 | /// \param __b16 |
3802 | /// An 8-bit integral value used to initialize bits [135:128] of the result. |
3803 | /// \param __b15 |
3804 | /// An 8-bit integral value used to initialize bits [127:120] of the result. |
3805 | /// \param __b14 |
3806 | /// An 8-bit integral value used to initialize bits [119:112] of the result. |
3807 | /// \param __b13 |
3808 | /// An 8-bit integral value used to initialize bits [111:104] of the result. |
3809 | /// \param __b12 |
3810 | /// An 8-bit integral value used to initialize bits [103:96] of the result. |
3811 | /// \param __b11 |
3812 | /// An 8-bit integral value used to initialize bits [95:88] of the result. |
3813 | /// \param __b10 |
3814 | /// An 8-bit integral value used to initialize bits [87:80] of the result. |
3815 | /// \param __b09 |
3816 | /// An 8-bit integral value used to initialize bits [79:72] of the result. |
3817 | /// \param __b08 |
3818 | /// An 8-bit integral value used to initialize bits [71:64] of the result. |
3819 | /// \param __b07 |
3820 | /// An 8-bit integral value used to initialize bits [63:56] of the result. |
3821 | /// \param __b06 |
3822 | /// An 8-bit integral value used to initialize bits [55:48] of the result. |
3823 | /// \param __b05 |
3824 | /// An 8-bit integral value used to initialize bits [47:40] of the result. |
3825 | /// \param __b04 |
3826 | /// An 8-bit integral value used to initialize bits [39:32] of the result. |
3827 | /// \param __b03 |
3828 | /// An 8-bit integral value used to initialize bits [31:24] of the result. |
3829 | /// \param __b02 |
3830 | /// An 8-bit integral value used to initialize bits [23:16] of the result. |
3831 | /// \param __b01 |
3832 | /// An 8-bit integral value used to initialize bits [15:8] of the result. |
3833 | /// \param __b00 |
3834 | /// An 8-bit integral value used to initialize bits [7:0] of the result. |
3835 | /// \returns An initialized 256-bit integer vector. |
3836 | static __inline __m256i __DEFAULT_FN_ATTRS |
3837 | _mm256_set_epi8(char __b31, char __b30, char __b29, char __b28, |
3838 | char __b27, char __b26, char __b25, char __b24, |
3839 | char __b23, char __b22, char __b21, char __b20, |
3840 | char __b19, char __b18, char __b17, char __b16, |
3841 | char __b15, char __b14, char __b13, char __b12, |
3842 | char __b11, char __b10, char __b09, char __b08, |
3843 | char __b07, char __b06, char __b05, char __b04, |
3844 | char __b03, char __b02, char __b01, char __b00) |
3845 | { |
3846 | return __extension__ (__m256i)(__v32qi){ |
3847 | __b00, __b01, __b02, __b03, __b04, __b05, __b06, __b07, |
3848 | __b08, __b09, __b10, __b11, __b12, __b13, __b14, __b15, |
3849 | __b16, __b17, __b18, __b19, __b20, __b21, __b22, __b23, |
3850 | __b24, __b25, __b26, __b27, __b28, __b29, __b30, __b31 |
3851 | }; |
3852 | } |
3853 | |
3854 | /// Constructs a 256-bit integer vector initialized with the specified |
3855 | /// 64-bit integral values. |
3856 | /// |
3857 | /// \headerfile <x86intrin.h> |
3858 | /// |
3859 | /// This intrinsic corresponds to the <c> VPUNPCKLQDQ+VINSERTF128 </c> |
3860 | /// instruction. |
3861 | /// |
3862 | /// \param __a |
3863 | /// A 64-bit integral value used to initialize bits [255:192] of the result. |
3864 | /// \param __b |
3865 | /// A 64-bit integral value used to initialize bits [191:128] of the result. |
3866 | /// \param __c |
3867 | /// A 64-bit integral value used to initialize bits [127:64] of the result. |
3868 | /// \param __d |
3869 | /// A 64-bit integral value used to initialize bits [63:0] of the result. |
3870 | /// \returns An initialized 256-bit integer vector. |
3871 | static __inline __m256i __DEFAULT_FN_ATTRS |
3872 | _mm256_set_epi64x(long long __a, long long __b, long long __c, long long __d) |
3873 | { |
3874 | return __extension__ (__m256i)(__v4di){ __d, __c, __b, __a }; |
3875 | } |
3876 | |
3877 | /* Create vectors with elements in reverse order */ |
3878 | /// Constructs a 256-bit floating-point vector of [4 x double], |
3879 | /// initialized in reverse order with the specified double-precision |
3880 | /// floating-point values. |
3881 | /// |
3882 | /// \headerfile <x86intrin.h> |
3883 | /// |
3884 | /// This intrinsic corresponds to the <c> VUNPCKLPD+VINSERTF128 </c> |
3885 | /// instruction. |
3886 | /// |
3887 | /// \param __a |
3888 | /// A double-precision floating-point value used to initialize bits [63:0] |
3889 | /// of the result. |
3890 | /// \param __b |
3891 | /// A double-precision floating-point value used to initialize bits [127:64] |
3892 | /// of the result. |
3893 | /// \param __c |
3894 | /// A double-precision floating-point value used to initialize bits [191:128] |
3895 | /// of the result. |
3896 | /// \param __d |
3897 | /// A double-precision floating-point value used to initialize bits [255:192] |
3898 | /// of the result. |
3899 | /// \returns An initialized 256-bit floating-point vector of [4 x double]. |
3900 | static __inline __m256d __DEFAULT_FN_ATTRS |
3901 | _mm256_setr_pd(double __a, double __b, double __c, double __d) |
3902 | { |
3903 | return _mm256_set_pd(__d, __c, __b, __a); |
3904 | } |
3905 | |
3906 | /// Constructs a 256-bit floating-point vector of [8 x float], |
3907 | /// initialized in reverse order with the specified single-precision |
3908 | /// float-point values. |
3909 | /// |
3910 | /// \headerfile <x86intrin.h> |
3911 | /// |
3912 | /// This intrinsic is a utility function and does not correspond to a specific |
3913 | /// instruction. |
3914 | /// |
3915 | /// \param __a |
3916 | /// A single-precision floating-point value used to initialize bits [31:0] |
3917 | /// of the result. |
3918 | /// \param __b |
3919 | /// A single-precision floating-point value used to initialize bits [63:32] |
3920 | /// of the result. |
3921 | /// \param __c |
3922 | /// A single-precision floating-point value used to initialize bits [95:64] |
3923 | /// of the result. |
3924 | /// \param __d |
3925 | /// A single-precision floating-point value used to initialize bits [127:96] |
3926 | /// of the result. |
3927 | /// \param __e |
3928 | /// A single-precision floating-point value used to initialize bits [159:128] |
3929 | /// of the result. |
3930 | /// \param __f |
3931 | /// A single-precision floating-point value used to initialize bits [191:160] |
3932 | /// of the result. |
3933 | /// \param __g |
3934 | /// A single-precision floating-point value used to initialize bits [223:192] |
3935 | /// of the result. |
3936 | /// \param __h |
3937 | /// A single-precision floating-point value used to initialize bits [255:224] |
3938 | /// of the result. |
3939 | /// \returns An initialized 256-bit floating-point vector of [8 x float]. |
3940 | static __inline __m256 __DEFAULT_FN_ATTRS |
3941 | _mm256_setr_ps(float __a, float __b, float __c, float __d, |
3942 | float __e, float __f, float __g, float __h) |
3943 | { |
3944 | return _mm256_set_ps(__h, __g, __f, __e, __d, __c, __b, __a); |
3945 | } |
3946 | |
3947 | /// Constructs a 256-bit integer vector, initialized in reverse order |
3948 | /// with the specified 32-bit integral values. |
3949 | /// |
3950 | /// \headerfile <x86intrin.h> |
3951 | /// |
3952 | /// This intrinsic is a utility function and does not correspond to a specific |
3953 | /// instruction. |
3954 | /// |
3955 | /// \param __i0 |
3956 | /// A 32-bit integral value used to initialize bits [31:0] of the result. |
3957 | /// \param __i1 |
3958 | /// A 32-bit integral value used to initialize bits [63:32] of the result. |
3959 | /// \param __i2 |
3960 | /// A 32-bit integral value used to initialize bits [95:64] of the result. |
3961 | /// \param __i3 |
3962 | /// A 32-bit integral value used to initialize bits [127:96] of the result. |
3963 | /// \param __i4 |
3964 | /// A 32-bit integral value used to initialize bits [159:128] of the result. |
3965 | /// \param __i5 |
3966 | /// A 32-bit integral value used to initialize bits [191:160] of the result. |
3967 | /// \param __i6 |
3968 | /// A 32-bit integral value used to initialize bits [223:192] of the result. |
3969 | /// \param __i7 |
3970 | /// A 32-bit integral value used to initialize bits [255:224] of the result. |
3971 | /// \returns An initialized 256-bit integer vector. |
3972 | static __inline __m256i __DEFAULT_FN_ATTRS |
3973 | _mm256_setr_epi32(int __i0, int __i1, int __i2, int __i3, |
3974 | int __i4, int __i5, int __i6, int __i7) |
3975 | { |
3976 | return _mm256_set_epi32(__i7, __i6, __i5, __i4, __i3, __i2, __i1, __i0); |
3977 | } |
3978 | |
3979 | /// Constructs a 256-bit integer vector, initialized in reverse order |
3980 | /// with the specified 16-bit integral values. |
3981 | /// |
3982 | /// \headerfile <x86intrin.h> |
3983 | /// |
3984 | /// This intrinsic is a utility function and does not correspond to a specific |
3985 | /// instruction. |
3986 | /// |
3987 | /// \param __w15 |
3988 | /// A 16-bit integral value used to initialize bits [15:0] of the result. |
3989 | /// \param __w14 |
3990 | /// A 16-bit integral value used to initialize bits [31:16] of the result. |
3991 | /// \param __w13 |
3992 | /// A 16-bit integral value used to initialize bits [47:32] of the result. |
3993 | /// \param __w12 |
3994 | /// A 16-bit integral value used to initialize bits [63:48] of the result. |
3995 | /// \param __w11 |
3996 | /// A 16-bit integral value used to initialize bits [79:64] of the result. |
3997 | /// \param __w10 |
3998 | /// A 16-bit integral value used to initialize bits [95:80] of the result. |
3999 | /// \param __w09 |
4000 | /// A 16-bit integral value used to initialize bits [111:96] of the result. |
4001 | /// \param __w08 |
4002 | /// A 16-bit integral value used to initialize bits [127:112] of the result. |
4003 | /// \param __w07 |
4004 | /// A 16-bit integral value used to initialize bits [143:128] of the result. |
4005 | /// \param __w06 |
4006 | /// A 16-bit integral value used to initialize bits [159:144] of the result. |
4007 | /// \param __w05 |
4008 | /// A 16-bit integral value used to initialize bits [175:160] of the result. |
4009 | /// \param __w04 |
4010 | /// A 16-bit integral value used to initialize bits [191:176] of the result. |
4011 | /// \param __w03 |
4012 | /// A 16-bit integral value used to initialize bits [207:192] of the result. |
4013 | /// \param __w02 |
4014 | /// A 16-bit integral value used to initialize bits [223:208] of the result. |
4015 | /// \param __w01 |
4016 | /// A 16-bit integral value used to initialize bits [239:224] of the result. |
4017 | /// \param __w00 |
4018 | /// A 16-bit integral value used to initialize bits [255:240] of the result. |
4019 | /// \returns An initialized 256-bit integer vector. |
4020 | static __inline __m256i __DEFAULT_FN_ATTRS |
4021 | _mm256_setr_epi16(short __w15, short __w14, short __w13, short __w12, |
4022 | short __w11, short __w10, short __w09, short __w08, |
4023 | short __w07, short __w06, short __w05, short __w04, |
4024 | short __w03, short __w02, short __w01, short __w00) |
4025 | { |
4026 | return _mm256_set_epi16(__w00, __w01, __w02, __w03, |
4027 | __w04, __w05, __w06, __w07, |
4028 | __w08, __w09, __w10, __w11, |
4029 | __w12, __w13, __w14, __w15); |
4030 | } |
4031 | |
4032 | /// Constructs a 256-bit integer vector, initialized in reverse order |
4033 | /// with the specified 8-bit integral values. |
4034 | /// |
4035 | /// \headerfile <x86intrin.h> |
4036 | /// |
4037 | /// This intrinsic is a utility function and does not correspond to a specific |
4038 | /// instruction. |
4039 | /// |
4040 | /// \param __b31 |
4041 | /// An 8-bit integral value used to initialize bits [7:0] of the result. |
4042 | /// \param __b30 |
4043 | /// An 8-bit integral value used to initialize bits [15:8] of the result. |
4044 | /// \param __b29 |
4045 | /// An 8-bit integral value used to initialize bits [23:16] of the result. |
4046 | /// \param __b28 |
4047 | /// An 8-bit integral value used to initialize bits [31:24] of the result. |
4048 | /// \param __b27 |
4049 | /// An 8-bit integral value used to initialize bits [39:32] of the result. |
4050 | /// \param __b26 |
4051 | /// An 8-bit integral value used to initialize bits [47:40] of the result. |
4052 | /// \param __b25 |
4053 | /// An 8-bit integral value used to initialize bits [55:48] of the result. |
4054 | /// \param __b24 |
4055 | /// An 8-bit integral value used to initialize bits [63:56] of the result. |
4056 | /// \param __b23 |
4057 | /// An 8-bit integral value used to initialize bits [71:64] of the result. |
4058 | /// \param __b22 |
4059 | /// An 8-bit integral value used to initialize bits [79:72] of the result. |
4060 | /// \param __b21 |
4061 | /// An 8-bit integral value used to initialize bits [87:80] of the result. |
4062 | /// \param __b20 |
4063 | /// An 8-bit integral value used to initialize bits [95:88] of the result. |
4064 | /// \param __b19 |
4065 | /// An 8-bit integral value used to initialize bits [103:96] of the result. |
4066 | /// \param __b18 |
4067 | /// An 8-bit integral value used to initialize bits [111:104] of the result. |
4068 | /// \param __b17 |
4069 | /// An 8-bit integral value used to initialize bits [119:112] of the result. |
4070 | /// \param __b16 |
4071 | /// An 8-bit integral value used to initialize bits [127:120] of the result. |
4072 | /// \param __b15 |
4073 | /// An 8-bit integral value used to initialize bits [135:128] of the result. |
4074 | /// \param __b14 |
4075 | /// An 8-bit integral value used to initialize bits [143:136] of the result. |
4076 | /// \param __b13 |
4077 | /// An 8-bit integral value used to initialize bits [151:144] of the result. |
4078 | /// \param __b12 |
4079 | /// An 8-bit integral value used to initialize bits [159:152] of the result. |
4080 | /// \param __b11 |
4081 | /// An 8-bit integral value used to initialize bits [167:160] of the result. |
4082 | /// \param __b10 |
4083 | /// An 8-bit integral value used to initialize bits [175:168] of the result. |
4084 | /// \param __b09 |
4085 | /// An 8-bit integral value used to initialize bits [183:176] of the result. |
4086 | /// \param __b08 |
4087 | /// An 8-bit integral value used to initialize bits [191:184] of the result. |
4088 | /// \param __b07 |
4089 | /// An 8-bit integral value used to initialize bits [199:192] of the result. |
4090 | /// \param __b06 |
4091 | /// An 8-bit integral value used to initialize bits [207:200] of the result. |
4092 | /// \param __b05 |
4093 | /// An 8-bit integral value used to initialize bits [215:208] of the result. |
4094 | /// \param __b04 |
4095 | /// An 8-bit integral value used to initialize bits [223:216] of the result. |
4096 | /// \param __b03 |
4097 | /// An 8-bit integral value used to initialize bits [231:224] of the result. |
4098 | /// \param __b02 |
4099 | /// An 8-bit integral value used to initialize bits [239:232] of the result. |
4100 | /// \param __b01 |
4101 | /// An 8-bit integral value used to initialize bits [247:240] of the result. |
4102 | /// \param __b00 |
4103 | /// An 8-bit integral value used to initialize bits [255:248] of the result. |
4104 | /// \returns An initialized 256-bit integer vector. |
4105 | static __inline __m256i __DEFAULT_FN_ATTRS |
4106 | _mm256_setr_epi8(char __b31, char __b30, char __b29, char __b28, |
4107 | char __b27, char __b26, char __b25, char __b24, |
4108 | char __b23, char __b22, char __b21, char __b20, |
4109 | char __b19, char __b18, char __b17, char __b16, |
4110 | char __b15, char __b14, char __b13, char __b12, |
4111 | char __b11, char __b10, char __b09, char __b08, |
4112 | char __b07, char __b06, char __b05, char __b04, |
4113 | char __b03, char __b02, char __b01, char __b00) |
4114 | { |
4115 | return _mm256_set_epi8(__b00, __b01, __b02, __b03, __b04, __b05, __b06, __b07, |
4116 | __b08, __b09, __b10, __b11, __b12, __b13, __b14, __b15, |
4117 | __b16, __b17, __b18, __b19, __b20, __b21, __b22, __b23, |
4118 | __b24, __b25, __b26, __b27, __b28, __b29, __b30, __b31); |
4119 | } |
4120 | |
4121 | /// Constructs a 256-bit integer vector, initialized in reverse order |
4122 | /// with the specified 64-bit integral values. |
4123 | /// |
4124 | /// \headerfile <x86intrin.h> |
4125 | /// |
4126 | /// This intrinsic corresponds to the <c> VPUNPCKLQDQ+VINSERTF128 </c> |
4127 | /// instruction. |
4128 | /// |
4129 | /// \param __a |
4130 | /// A 64-bit integral value used to initialize bits [63:0] of the result. |
4131 | /// \param __b |
4132 | /// A 64-bit integral value used to initialize bits [127:64] of the result. |
4133 | /// \param __c |
4134 | /// A 64-bit integral value used to initialize bits [191:128] of the result. |
4135 | /// \param __d |
4136 | /// A 64-bit integral value used to initialize bits [255:192] of the result. |
4137 | /// \returns An initialized 256-bit integer vector. |
4138 | static __inline __m256i __DEFAULT_FN_ATTRS |
4139 | _mm256_setr_epi64x(long long __a, long long __b, long long __c, long long __d) |
4140 | { |
4141 | return _mm256_set_epi64x(__d, __c, __b, __a); |
4142 | } |
4143 | |
4144 | /* Create vectors with repeated elements */ |
4145 | /// Constructs a 256-bit floating-point vector of [4 x double], with each |
4146 | /// of the four double-precision floating-point vector elements set to the |
4147 | /// specified double-precision floating-point value. |
4148 | /// |
4149 | /// \headerfile <x86intrin.h> |
4150 | /// |
4151 | /// This intrinsic corresponds to the <c> VMOVDDUP+VINSERTF128 </c> instruction. |
4152 | /// |
4153 | /// \param __w |
4154 | /// A double-precision floating-point value used to initialize each vector |
4155 | /// element of the result. |
4156 | /// \returns An initialized 256-bit floating-point vector of [4 x double]. |
4157 | static __inline __m256d __DEFAULT_FN_ATTRS |
4158 | _mm256_set1_pd(double __w) |
4159 | { |
4160 | return _mm256_set_pd(__w, __w, __w, __w); |
4161 | } |
4162 | |
4163 | /// Constructs a 256-bit floating-point vector of [8 x float], with each |
4164 | /// of the eight single-precision floating-point vector elements set to the |
4165 | /// specified single-precision floating-point value. |
4166 | /// |
4167 | /// \headerfile <x86intrin.h> |
4168 | /// |
4169 | /// This intrinsic corresponds to the <c> VPERMILPS+VINSERTF128 </c> |
4170 | /// instruction. |
4171 | /// |
4172 | /// \param __w |
4173 | /// A single-precision floating-point value used to initialize each vector |
4174 | /// element of the result. |
4175 | /// \returns An initialized 256-bit floating-point vector of [8 x float]. |
4176 | static __inline __m256 __DEFAULT_FN_ATTRS |
4177 | _mm256_set1_ps(float __w) |
4178 | { |
4179 | return _mm256_set_ps(__w, __w, __w, __w, __w, __w, __w, __w); |
4180 | } |
4181 | |
4182 | /// Constructs a 256-bit integer vector of [8 x i32], with each of the |
4183 | /// 32-bit integral vector elements set to the specified 32-bit integral |
4184 | /// value. |
4185 | /// |
4186 | /// \headerfile <x86intrin.h> |
4187 | /// |
4188 | /// This intrinsic corresponds to the <c> VPERMILPS+VINSERTF128 </c> |
4189 | /// instruction. |
4190 | /// |
4191 | /// \param __i |
4192 | /// A 32-bit integral value used to initialize each vector element of the |
4193 | /// result. |
4194 | /// \returns An initialized 256-bit integer vector of [8 x i32]. |
4195 | static __inline __m256i __DEFAULT_FN_ATTRS |
4196 | _mm256_set1_epi32(int __i) |
4197 | { |
4198 | return _mm256_set_epi32(__i, __i, __i, __i, __i, __i, __i, __i); |
4199 | } |
4200 | |
4201 | /// Constructs a 256-bit integer vector of [16 x i16], with each of the |
4202 | /// 16-bit integral vector elements set to the specified 16-bit integral |
4203 | /// value. |
4204 | /// |
4205 | /// \headerfile <x86intrin.h> |
4206 | /// |
4207 | /// This intrinsic corresponds to the <c> VPSHUFB+VINSERTF128 </c> instruction. |
4208 | /// |
4209 | /// \param __w |
4210 | /// A 16-bit integral value used to initialize each vector element of the |
4211 | /// result. |
4212 | /// \returns An initialized 256-bit integer vector of [16 x i16]. |
4213 | static __inline __m256i __DEFAULT_FN_ATTRS |
4214 | _mm256_set1_epi16(short __w) |
4215 | { |
4216 | return _mm256_set_epi16(__w, __w, __w, __w, __w, __w, __w, __w, |
4217 | __w, __w, __w, __w, __w, __w, __w, __w); |
4218 | } |
4219 | |
4220 | /// Constructs a 256-bit integer vector of [32 x i8], with each of the |
4221 | /// 8-bit integral vector elements set to the specified 8-bit integral value. |
4222 | /// |
4223 | /// \headerfile <x86intrin.h> |
4224 | /// |
4225 | /// This intrinsic corresponds to the <c> VPSHUFB+VINSERTF128 </c> instruction. |
4226 | /// |
4227 | /// \param __b |
4228 | /// An 8-bit integral value used to initialize each vector element of the |
4229 | /// result. |
4230 | /// \returns An initialized 256-bit integer vector of [32 x i8]. |
4231 | static __inline __m256i __DEFAULT_FN_ATTRS |
4232 | _mm256_set1_epi8(char __b) |
4233 | { |
4234 | return _mm256_set_epi8(__b, __b, __b, __b, __b, __b, __b, __b, |
4235 | __b, __b, __b, __b, __b, __b, __b, __b, |
4236 | __b, __b, __b, __b, __b, __b, __b, __b, |
4237 | __b, __b, __b, __b, __b, __b, __b, __b); |
4238 | } |
4239 | |
4240 | /// Constructs a 256-bit integer vector of [4 x i64], with each of the |
4241 | /// 64-bit integral vector elements set to the specified 64-bit integral |
4242 | /// value. |
4243 | /// |
4244 | /// \headerfile <x86intrin.h> |
4245 | /// |
4246 | /// This intrinsic corresponds to the <c> VMOVDDUP+VINSERTF128 </c> instruction. |
4247 | /// |
4248 | /// \param __q |
4249 | /// A 64-bit integral value used to initialize each vector element of the |
4250 | /// result. |
4251 | /// \returns An initialized 256-bit integer vector of [4 x i64]. |
4252 | static __inline __m256i __DEFAULT_FN_ATTRS |
4253 | _mm256_set1_epi64x(long long __q) |
4254 | { |
4255 | return _mm256_set_epi64x(__q, __q, __q, __q); |
4256 | } |
4257 | |
4258 | /* Create __zeroed vectors */ |
4259 | /// Constructs a 256-bit floating-point vector of [4 x double] with all |
4260 | /// vector elements initialized to zero. |
4261 | /// |
4262 | /// \headerfile <x86intrin.h> |
4263 | /// |
4264 | /// This intrinsic corresponds to the <c> VXORPS </c> instruction. |
4265 | /// |
4266 | /// \returns A 256-bit vector of [4 x double] with all elements set to zero. |
4267 | static __inline __m256d __DEFAULT_FN_ATTRS |
4268 | _mm256_setzero_pd(void) |
4269 | { |
4270 | return __extension__ (__m256d){ 0, 0, 0, 0 }; |
4271 | } |
4272 | |
4273 | /// Constructs a 256-bit floating-point vector of [8 x float] with all |
4274 | /// vector elements initialized to zero. |
4275 | /// |
4276 | /// \headerfile <x86intrin.h> |
4277 | /// |
4278 | /// This intrinsic corresponds to the <c> VXORPS </c> instruction. |
4279 | /// |
4280 | /// \returns A 256-bit vector of [8 x float] with all elements set to zero. |
4281 | static __inline __m256 __DEFAULT_FN_ATTRS |
4282 | _mm256_setzero_ps(void) |
4283 | { |
4284 | return __extension__ (__m256){ 0, 0, 0, 0, 0, 0, 0, 0 }; |
4285 | } |
4286 | |
4287 | /// Constructs a 256-bit integer vector initialized to zero. |
4288 | /// |
4289 | /// \headerfile <x86intrin.h> |
4290 | /// |
4291 | /// This intrinsic corresponds to the <c> VXORPS </c> instruction. |
4292 | /// |
4293 | /// \returns A 256-bit integer vector initialized to zero. |
4294 | static __inline __m256i __DEFAULT_FN_ATTRS |
4295 | _mm256_setzero_si256(void) |
4296 | { |
4297 | return __extension__ (__m256i)(__v4di){ 0, 0, 0, 0 }; |
4298 | } |
4299 | |
4300 | /* Cast between vector types */ |
4301 | /// Casts a 256-bit floating-point vector of [4 x double] into a 256-bit |
4302 | /// floating-point vector of [8 x float]. |
4303 | /// |
4304 | /// \headerfile <x86intrin.h> |
4305 | /// |
4306 | /// This intrinsic has no corresponding instruction. |
4307 | /// |
4308 | /// \param __a |
4309 | /// A 256-bit floating-point vector of [4 x double]. |
4310 | /// \returns A 256-bit floating-point vector of [8 x float] containing the same |
4311 | /// bitwise pattern as the parameter. |
4312 | static __inline __m256 __DEFAULT_FN_ATTRS |
4313 | _mm256_castpd_ps(__m256d __a) |
4314 | { |
4315 | return (__m256)__a; |
4316 | } |
4317 | |
4318 | /// Casts a 256-bit floating-point vector of [4 x double] into a 256-bit |
4319 | /// integer vector. |
4320 | /// |
4321 | /// \headerfile <x86intrin.h> |
4322 | /// |
4323 | /// This intrinsic has no corresponding instruction. |
4324 | /// |
4325 | /// \param __a |
4326 | /// A 256-bit floating-point vector of [4 x double]. |
4327 | /// \returns A 256-bit integer vector containing the same bitwise pattern as the |
4328 | /// parameter. |
4329 | static __inline __m256i __DEFAULT_FN_ATTRS |
4330 | _mm256_castpd_si256(__m256d __a) |
4331 | { |
4332 | return (__m256i)__a; |
4333 | } |
4334 | |
4335 | /// Casts a 256-bit floating-point vector of [8 x float] into a 256-bit |
4336 | /// floating-point vector of [4 x double]. |
4337 | /// |
4338 | /// \headerfile <x86intrin.h> |
4339 | /// |
4340 | /// This intrinsic has no corresponding instruction. |
4341 | /// |
4342 | /// \param __a |
4343 | /// A 256-bit floating-point vector of [8 x float]. |
4344 | /// \returns A 256-bit floating-point vector of [4 x double] containing the same |
4345 | /// bitwise pattern as the parameter. |
4346 | static __inline __m256d __DEFAULT_FN_ATTRS |
4347 | _mm256_castps_pd(__m256 __a) |
4348 | { |
4349 | return (__m256d)__a; |
4350 | } |
4351 | |
4352 | /// Casts a 256-bit floating-point vector of [8 x float] into a 256-bit |
4353 | /// integer vector. |
4354 | /// |
4355 | /// \headerfile <x86intrin.h> |
4356 | /// |
4357 | /// This intrinsic has no corresponding instruction. |
4358 | /// |
4359 | /// \param __a |
4360 | /// A 256-bit floating-point vector of [8 x float]. |
4361 | /// \returns A 256-bit integer vector containing the same bitwise pattern as the |
4362 | /// parameter. |
4363 | static __inline __m256i __DEFAULT_FN_ATTRS |
4364 | _mm256_castps_si256(__m256 __a) |
4365 | { |
4366 | return (__m256i)__a; |
4367 | } |
4368 | |
4369 | /// Casts a 256-bit integer vector into a 256-bit floating-point vector |
4370 | /// of [8 x float]. |
4371 | /// |
4372 | /// \headerfile <x86intrin.h> |
4373 | /// |
4374 | /// This intrinsic has no corresponding instruction. |
4375 | /// |
4376 | /// \param __a |
4377 | /// A 256-bit integer vector. |
4378 | /// \returns A 256-bit floating-point vector of [8 x float] containing the same |
4379 | /// bitwise pattern as the parameter. |
4380 | static __inline __m256 __DEFAULT_FN_ATTRS |
4381 | _mm256_castsi256_ps(__m256i __a) |
4382 | { |
4383 | return (__m256)__a; |
4384 | } |
4385 | |
4386 | /// Casts a 256-bit integer vector into a 256-bit floating-point vector |
4387 | /// of [4 x double]. |
4388 | /// |
4389 | /// \headerfile <x86intrin.h> |
4390 | /// |
4391 | /// This intrinsic has no corresponding instruction. |
4392 | /// |
4393 | /// \param __a |
4394 | /// A 256-bit integer vector. |
4395 | /// \returns A 256-bit floating-point vector of [4 x double] containing the same |
4396 | /// bitwise pattern as the parameter. |
4397 | static __inline __m256d __DEFAULT_FN_ATTRS |
4398 | _mm256_castsi256_pd(__m256i __a) |
4399 | { |
4400 | return (__m256d)__a; |
4401 | } |
4402 | |
4403 | /// Returns the lower 128 bits of a 256-bit floating-point vector of |
4404 | /// [4 x double] as a 128-bit floating-point vector of [2 x double]. |
4405 | /// |
4406 | /// \headerfile <x86intrin.h> |
4407 | /// |
4408 | /// This intrinsic has no corresponding instruction. |
4409 | /// |
4410 | /// \param __a |
4411 | /// A 256-bit floating-point vector of [4 x double]. |
4412 | /// \returns A 128-bit floating-point vector of [2 x double] containing the |
4413 | /// lower 128 bits of the parameter. |
4414 | static __inline __m128d __DEFAULT_FN_ATTRS |
4415 | _mm256_castpd256_pd128(__m256d __a) |
4416 | { |
4417 | return __builtin_shufflevector((__v4df)__a, (__v4df)__a, 0, 1); |
4418 | } |
4419 | |
4420 | /// Returns the lower 128 bits of a 256-bit floating-point vector of |
4421 | /// [8 x float] as a 128-bit floating-point vector of [4 x float]. |
4422 | /// |
4423 | /// \headerfile <x86intrin.h> |
4424 | /// |
4425 | /// This intrinsic has no corresponding instruction. |
4426 | /// |
4427 | /// \param __a |
4428 | /// A 256-bit floating-point vector of [8 x float]. |
4429 | /// \returns A 128-bit floating-point vector of [4 x float] containing the |
4430 | /// lower 128 bits of the parameter. |
4431 | static __inline __m128 __DEFAULT_FN_ATTRS |
4432 | _mm256_castps256_ps128(__m256 __a) |
4433 | { |
4434 | return __builtin_shufflevector((__v8sf)__a, (__v8sf)__a, 0, 1, 2, 3); |
4435 | } |
4436 | |
4437 | /// Truncates a 256-bit integer vector into a 128-bit integer vector. |
4438 | /// |
4439 | /// \headerfile <x86intrin.h> |
4440 | /// |
4441 | /// This intrinsic has no corresponding instruction. |
4442 | /// |
4443 | /// \param __a |
4444 | /// A 256-bit integer vector. |
4445 | /// \returns A 128-bit integer vector containing the lower 128 bits of the |
4446 | /// parameter. |
4447 | static __inline __m128i __DEFAULT_FN_ATTRS |
4448 | _mm256_castsi256_si128(__m256i __a) |
4449 | { |
4450 | return __builtin_shufflevector((__v4di)__a, (__v4di)__a, 0, 1); |
4451 | } |
4452 | |
4453 | /// Constructs a 256-bit floating-point vector of [4 x double] from a |
4454 | /// 128-bit floating-point vector of [2 x double]. |
4455 | /// |
4456 | /// The lower 128 bits contain the value of the source vector. The contents |
4457 | /// of the upper 128 bits are undefined. |
4458 | /// |
4459 | /// \headerfile <x86intrin.h> |
4460 | /// |
4461 | /// This intrinsic has no corresponding instruction. |
4462 | /// |
4463 | /// \param __a |
4464 | /// A 128-bit vector of [2 x double]. |
4465 | /// \returns A 256-bit floating-point vector of [4 x double]. The lower 128 bits |
4466 | /// contain the value of the parameter. The contents of the upper 128 bits |
4467 | /// are undefined. |
4468 | static __inline __m256d __DEFAULT_FN_ATTRS |
4469 | _mm256_castpd128_pd256(__m128d __a) |
4470 | { |
4471 | return __builtin_shufflevector((__v2df)__a, (__v2df)__a, 0, 1, -1, -1); |
4472 | } |
4473 | |
4474 | /// Constructs a 256-bit floating-point vector of [8 x float] from a |
4475 | /// 128-bit floating-point vector of [4 x float]. |
4476 | /// |
4477 | /// The lower 128 bits contain the value of the source vector. The contents |
4478 | /// of the upper 128 bits are undefined. |
4479 | /// |
4480 | /// \headerfile <x86intrin.h> |
4481 | /// |
4482 | /// This intrinsic has no corresponding instruction. |
4483 | /// |
4484 | /// \param __a |
4485 | /// A 128-bit vector of [4 x float]. |
4486 | /// \returns A 256-bit floating-point vector of [8 x float]. The lower 128 bits |
4487 | /// contain the value of the parameter. The contents of the upper 128 bits |
4488 | /// are undefined. |
4489 | static __inline __m256 __DEFAULT_FN_ATTRS |
4490 | _mm256_castps128_ps256(__m128 __a) |
4491 | { |
4492 | return __builtin_shufflevector((__v4sf)__a, (__v4sf)__a, 0, 1, 2, 3, -1, -1, -1, -1); |
4493 | } |
4494 | |
4495 | /// Constructs a 256-bit integer vector from a 128-bit integer vector. |
4496 | /// |
4497 | /// The lower 128 bits contain the value of the source vector. The contents |
4498 | /// of the upper 128 bits are undefined. |
4499 | /// |
4500 | /// \headerfile <x86intrin.h> |
4501 | /// |
4502 | /// This intrinsic has no corresponding instruction. |
4503 | /// |
4504 | /// \param __a |
4505 | /// A 128-bit integer vector. |
4506 | /// \returns A 256-bit integer vector. The lower 128 bits contain the value of |
4507 | /// the parameter. The contents of the upper 128 bits are undefined. |
4508 | static __inline __m256i __DEFAULT_FN_ATTRS |
4509 | _mm256_castsi128_si256(__m128i __a) |
4510 | { |
4511 | return __builtin_shufflevector((__v2di)__a, (__v2di)__a, 0, 1, -1, -1); |
4512 | } |
4513 | |
4514 | /// Constructs a 256-bit floating-point vector of [4 x double] from a |
4515 | /// 128-bit floating-point vector of [2 x double]. The lower 128 bits |
4516 | /// contain the value of the source vector. The upper 128 bits are set |
4517 | /// to zero. |
4518 | /// |
4519 | /// \headerfile <x86intrin.h> |
4520 | /// |
4521 | /// This intrinsic has no corresponding instruction. |
4522 | /// |
4523 | /// \param __a |
4524 | /// A 128-bit vector of [2 x double]. |
4525 | /// \returns A 256-bit floating-point vector of [4 x double]. The lower 128 bits |
4526 | /// contain the value of the parameter. The upper 128 bits are set to zero. |
4527 | static __inline __m256d __DEFAULT_FN_ATTRS |
4528 | _mm256_zextpd128_pd256(__m128d __a) |
4529 | { |
4530 | return __builtin_shufflevector((__v2df)__a, (__v2df)_mm_setzero_pd(), 0, 1, 2, 3); |
4531 | } |
4532 | |
4533 | /// Constructs a 256-bit floating-point vector of [8 x float] from a |
4534 | /// 128-bit floating-point vector of [4 x float]. The lower 128 bits contain |
4535 | /// the value of the source vector. The upper 128 bits are set to zero. |
4536 | /// |
4537 | /// \headerfile <x86intrin.h> |
4538 | /// |
4539 | /// This intrinsic has no corresponding instruction. |
4540 | /// |
4541 | /// \param __a |
4542 | /// A 128-bit vector of [4 x float]. |
4543 | /// \returns A 256-bit floating-point vector of [8 x float]. The lower 128 bits |
4544 | /// contain the value of the parameter. The upper 128 bits are set to zero. |
4545 | static __inline __m256 __DEFAULT_FN_ATTRS |
4546 | _mm256_zextps128_ps256(__m128 __a) |
4547 | { |
4548 | return __builtin_shufflevector((__v4sf)__a, (__v4sf)_mm_setzero_ps(), 0, 1, 2, 3, 4, 5, 6, 7); |
4549 | } |
4550 | |
4551 | /// Constructs a 256-bit integer vector from a 128-bit integer vector. |
4552 | /// The lower 128 bits contain the value of the source vector. The upper |
4553 | /// 128 bits are set to zero. |
4554 | /// |
4555 | /// \headerfile <x86intrin.h> |
4556 | /// |
4557 | /// This intrinsic has no corresponding instruction. |
4558 | /// |
4559 | /// \param __a |
4560 | /// A 128-bit integer vector. |
4561 | /// \returns A 256-bit integer vector. The lower 128 bits contain the value of |
4562 | /// the parameter. The upper 128 bits are set to zero. |
4563 | static __inline __m256i __DEFAULT_FN_ATTRS |
4564 | _mm256_zextsi128_si256(__m128i __a) |
4565 | { |
4566 | return __builtin_shufflevector((__v2di)__a, (__v2di)_mm_setzero_si128(), 0, 1, 2, 3); |
4567 | } |
4568 | |
4569 | /* |
4570 | Vector insert. |
4571 | We use macros rather than inlines because we only want to accept |
4572 | invocations where the immediate M is a constant expression. |
4573 | */ |
4574 | /// Constructs a new 256-bit vector of [8 x float] by first duplicating |
4575 | /// a 256-bit vector of [8 x float] given in the first parameter, and then |
4576 | /// replacing either the upper or the lower 128 bits with the contents of a |
4577 | /// 128-bit vector of [4 x float] in the second parameter. |
4578 | /// |
4579 | /// The immediate integer parameter determines between the upper or the lower |
4580 | /// 128 bits. |
4581 | /// |
4582 | /// \headerfile <x86intrin.h> |
4583 | /// |
4584 | /// \code |
4585 | /// __m256 _mm256_insertf128_ps(__m256 V1, __m128 V2, const int M); |
4586 | /// \endcode |
4587 | /// |
4588 | /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction. |
4589 | /// |
4590 | /// \param V1 |
4591 | /// A 256-bit vector of [8 x float]. This vector is copied to the result |
4592 | /// first, and then either the upper or the lower 128 bits of the result will |
4593 | /// be replaced by the contents of \a V2. |
4594 | /// \param V2 |
4595 | /// A 128-bit vector of [4 x float]. The contents of this parameter are |
4596 | /// written to either the upper or the lower 128 bits of the result depending |
4597 | /// on the value of parameter \a M. |
4598 | /// \param M |
4599 | /// An immediate integer. The least significant bit determines how the values |
4600 | /// from the two parameters are interleaved: \n |
4601 | /// If bit [0] of \a M is 0, \a V2 are copied to bits [127:0] of the result, |
4602 | /// and bits [255:128] of \a V1 are copied to bits [255:128] of the |
4603 | /// result. \n |
4604 | /// If bit [0] of \a M is 1, \a V2 are copied to bits [255:128] of the |
4605 | /// result, and bits [127:0] of \a V1 are copied to bits [127:0] of the |
4606 | /// result. |
4607 | /// \returns A 256-bit vector of [8 x float] containing the interleaved values. |
4608 | #define _mm256_insertf128_ps(V1, V2, M) \ |
4609 | (__m256)__builtin_ia32_vinsertf128_ps256((__v8sf)(__m256)(V1), \ |
4610 | (__v4sf)(__m128)(V2), (int)(M)) |
4611 | |
4612 | /// Constructs a new 256-bit vector of [4 x double] by first duplicating |
4613 | /// a 256-bit vector of [4 x double] given in the first parameter, and then |
4614 | /// replacing either the upper or the lower 128 bits with the contents of a |
4615 | /// 128-bit vector of [2 x double] in the second parameter. |
4616 | /// |
4617 | /// The immediate integer parameter determines between the upper or the lower |
4618 | /// 128 bits. |
4619 | /// |
4620 | /// \headerfile <x86intrin.h> |
4621 | /// |
4622 | /// \code |
4623 | /// __m256d _mm256_insertf128_pd(__m256d V1, __m128d V2, const int M); |
4624 | /// \endcode |
4625 | /// |
4626 | /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction. |
4627 | /// |
4628 | /// \param V1 |
4629 | /// A 256-bit vector of [4 x double]. This vector is copied to the result |
4630 | /// first, and then either the upper or the lower 128 bits of the result will |
4631 | /// be replaced by the contents of \a V2. |
4632 | /// \param V2 |
4633 | /// A 128-bit vector of [2 x double]. The contents of this parameter are |
4634 | /// written to either the upper or the lower 128 bits of the result depending |
4635 | /// on the value of parameter \a M. |
4636 | /// \param M |
4637 | /// An immediate integer. The least significant bit determines how the values |
4638 | /// from the two parameters are interleaved: \n |
4639 | /// If bit [0] of \a M is 0, \a V2 are copied to bits [127:0] of the result, |
4640 | /// and bits [255:128] of \a V1 are copied to bits [255:128] of the |
4641 | /// result. \n |
4642 | /// If bit [0] of \a M is 1, \a V2 are copied to bits [255:128] of the |
4643 | /// result, and bits [127:0] of \a V1 are copied to bits [127:0] of the |
4644 | /// result. |
4645 | /// \returns A 256-bit vector of [4 x double] containing the interleaved values. |
4646 | #define _mm256_insertf128_pd(V1, V2, M) \ |
4647 | (__m256d)__builtin_ia32_vinsertf128_pd256((__v4df)(__m256d)(V1), \ |
4648 | (__v2df)(__m128d)(V2), (int)(M)) |
4649 | |
4650 | /// Constructs a new 256-bit integer vector by first duplicating a |
4651 | /// 256-bit integer vector given in the first parameter, and then replacing |
4652 | /// either the upper or the lower 128 bits with the contents of a 128-bit |
4653 | /// integer vector in the second parameter. |
4654 | /// |
4655 | /// The immediate integer parameter determines between the upper or the lower |
4656 | /// 128 bits. |
4657 | /// |
4658 | /// \headerfile <x86intrin.h> |
4659 | /// |
4660 | /// \code |
4661 | /// __m256i _mm256_insertf128_si256(__m256i V1, __m128i V2, const int M); |
4662 | /// \endcode |
4663 | /// |
4664 | /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction. |
4665 | /// |
4666 | /// \param V1 |
4667 | /// A 256-bit integer vector. This vector is copied to the result first, and |
4668 | /// then either the upper or the lower 128 bits of the result will be |
4669 | /// replaced by the contents of \a V2. |
4670 | /// \param V2 |
4671 | /// A 128-bit integer vector. The contents of this parameter are written to |
4672 | /// either the upper or the lower 128 bits of the result depending on the |
4673 | /// value of parameter \a M. |
4674 | /// \param M |
4675 | /// An immediate integer. The least significant bit determines how the values |
4676 | /// from the two parameters are interleaved: \n |
4677 | /// If bit [0] of \a M is 0, \a V2 are copied to bits [127:0] of the result, |
4678 | /// and bits [255:128] of \a V1 are copied to bits [255:128] of the |
4679 | /// result. \n |
4680 | /// If bit [0] of \a M is 1, \a V2 are copied to bits [255:128] of the |
4681 | /// result, and bits [127:0] of \a V1 are copied to bits [127:0] of the |
4682 | /// result. |
4683 | /// \returns A 256-bit integer vector containing the interleaved values. |
4684 | #define _mm256_insertf128_si256(V1, V2, M) \ |
4685 | (__m256i)__builtin_ia32_vinsertf128_si256((__v8si)(__m256i)(V1), \ |
4686 | (__v4si)(__m128i)(V2), (int)(M)) |
4687 | |
4688 | /* |
4689 | Vector extract. |
4690 | We use macros rather than inlines because we only want to accept |
4691 | invocations where the immediate M is a constant expression. |
4692 | */ |
4693 | /// Extracts either the upper or the lower 128 bits from a 256-bit vector |
4694 | /// of [8 x float], as determined by the immediate integer parameter, and |
4695 | /// returns the extracted bits as a 128-bit vector of [4 x float]. |
4696 | /// |
4697 | /// \headerfile <x86intrin.h> |
4698 | /// |
4699 | /// \code |
4700 | /// __m128 _mm256_extractf128_ps(__m256 V, const int M); |
4701 | /// \endcode |
4702 | /// |
4703 | /// This intrinsic corresponds to the <c> VEXTRACTF128 </c> instruction. |
4704 | /// |
4705 | /// \param V |
4706 | /// A 256-bit vector of [8 x float]. |
4707 | /// \param M |
4708 | /// An immediate integer. The least significant bit determines which bits are |
4709 | /// extracted from the first parameter: \n |
4710 | /// If bit [0] of \a M is 0, bits [127:0] of \a V are copied to the |
4711 | /// result. \n |
4712 | /// If bit [0] of \a M is 1, bits [255:128] of \a V are copied to the result. |
4713 | /// \returns A 128-bit vector of [4 x float] containing the extracted bits. |
4714 | #define _mm256_extractf128_ps(V, M) \ |
4715 | (__m128)__builtin_ia32_vextractf128_ps256((__v8sf)(__m256)(V), (int)(M)) |
4716 | |
4717 | /// Extracts either the upper or the lower 128 bits from a 256-bit vector |
4718 | /// of [4 x double], as determined by the immediate integer parameter, and |
4719 | /// returns the extracted bits as a 128-bit vector of [2 x double]. |
4720 | /// |
4721 | /// \headerfile <x86intrin.h> |
4722 | /// |
4723 | /// \code |
4724 | /// __m128d _mm256_extractf128_pd(__m256d V, const int M); |
4725 | /// \endcode |
4726 | /// |
4727 | /// This intrinsic corresponds to the <c> VEXTRACTF128 </c> instruction. |
4728 | /// |
4729 | /// \param V |
4730 | /// A 256-bit vector of [4 x double]. |
4731 | /// \param M |
4732 | /// An immediate integer. The least significant bit determines which bits are |
4733 | /// extracted from the first parameter: \n |
4734 | /// If bit [0] of \a M is 0, bits [127:0] of \a V are copied to the |
4735 | /// result. \n |
4736 | /// If bit [0] of \a M is 1, bits [255:128] of \a V are copied to the result. |
4737 | /// \returns A 128-bit vector of [2 x double] containing the extracted bits. |
4738 | #define _mm256_extractf128_pd(V, M) \ |
4739 | (__m128d)__builtin_ia32_vextractf128_pd256((__v4df)(__m256d)(V), (int)(M)) |
4740 | |
4741 | /// Extracts either the upper or the lower 128 bits from a 256-bit |
4742 | /// integer vector, as determined by the immediate integer parameter, and |
4743 | /// returns the extracted bits as a 128-bit integer vector. |
4744 | /// |
4745 | /// \headerfile <x86intrin.h> |
4746 | /// |
4747 | /// \code |
4748 | /// __m128i _mm256_extractf128_si256(__m256i 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 integer vector. |
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 integer vector containing the extracted bits. |
4762 | #define _mm256_extractf128_si256(V, M) \ |
4763 | (__m128i)__builtin_ia32_vextractf128_si256((__v8si)(__m256i)(V), (int)(M)) |
4764 | |
4765 | /* SIMD load ops (unaligned) */ |
4766 | /// Loads two 128-bit floating-point vectors of [4 x float] from |
4767 | /// unaligned memory locations and constructs a 256-bit floating-point vector |
4768 | /// of [8 x float] by concatenating the two 128-bit vectors. |
4769 | /// |
4770 | /// \headerfile <x86intrin.h> |
4771 | /// |
4772 | /// This intrinsic corresponds to load instructions followed by the |
4773 | /// <c> VINSERTF128 </c> instruction. |
4774 | /// |
4775 | /// \param __addr_hi |
4776 | /// A pointer to a 128-bit memory location containing 4 consecutive |
4777 | /// single-precision floating-point values. These values are to be copied to |
4778 | /// bits[255:128] of the result. The address of the memory location does not |
4779 | /// have to be aligned. |
4780 | /// \param __addr_lo |
4781 | /// A pointer to a 128-bit memory location containing 4 consecutive |
4782 | /// single-precision floating-point values. These values are to be copied to |
4783 | /// bits[127:0] of the result. The address of the memory location does not |
4784 | /// have to be aligned. |
4785 | /// \returns A 256-bit floating-point vector of [8 x float] containing the |
4786 | /// concatenated result. |
4787 | static __inline __m256 __DEFAULT_FN_ATTRS |
4788 | _mm256_loadu2_m128(float const *__addr_hi, float const *__addr_lo) |
4789 | { |
4790 | __m256 __v256 = _mm256_castps128_ps256(_mm_loadu_ps(__addr_lo)); |
4791 | return _mm256_insertf128_ps(__v256, _mm_loadu_ps(__addr_hi), 1); |
4792 | } |
4793 | |
4794 | /// Loads two 128-bit floating-point vectors of [2 x double] from |
4795 | /// unaligned memory locations and constructs a 256-bit floating-point vector |
4796 | /// of [4 x double] by concatenating the two 128-bit vectors. |
4797 | /// |
4798 | /// \headerfile <x86intrin.h> |
4799 | /// |
4800 | /// This intrinsic corresponds to load instructions followed by the |
4801 | /// <c> VINSERTF128 </c> instruction. |
4802 | /// |
4803 | /// \param __addr_hi |
4804 | /// A pointer to a 128-bit memory location containing two consecutive |
4805 | /// double-precision floating-point values. These values are to be copied to |
4806 | /// bits[255:128] of the result. The address of the memory location does not |
4807 | /// have to be aligned. |
4808 | /// \param __addr_lo |
4809 | /// A pointer to a 128-bit memory location containing two consecutive |
4810 | /// double-precision floating-point values. These values are to be copied to |
4811 | /// bits[127:0] of the result. The address of the memory location does not |
4812 | /// have to be aligned. |
4813 | /// \returns A 256-bit floating-point vector of [4 x double] containing the |
4814 | /// concatenated result. |
4815 | static __inline __m256d __DEFAULT_FN_ATTRS |
4816 | _mm256_loadu2_m128d(double const *__addr_hi, double const *__addr_lo) |
4817 | { |
4818 | __m256d __v256 = _mm256_castpd128_pd256(_mm_loadu_pd(__addr_lo)); |
4819 | return _mm256_insertf128_pd(__v256, _mm_loadu_pd(__addr_hi), 1); |
4820 | } |
4821 | |
4822 | /// Loads two 128-bit integer vectors from unaligned memory locations and |
4823 | /// constructs a 256-bit integer vector by concatenating the two 128-bit |
4824 | /// vectors. |
4825 | /// |
4826 | /// \headerfile <x86intrin.h> |
4827 | /// |
4828 | /// This intrinsic corresponds to load instructions followed by the |
4829 | /// <c> VINSERTF128 </c> instruction. |
4830 | /// |
4831 | /// \param __addr_hi |
4832 | /// A pointer to a 128-bit memory location containing a 128-bit integer |
4833 | /// vector. This vector is to be copied to bits[255:128] of the result. The |
4834 | /// address of the memory location does not have to be aligned. |
4835 | /// \param __addr_lo |
4836 | /// A pointer to a 128-bit memory location containing a 128-bit integer |
4837 | /// vector. This vector is to be copied to bits[127:0] of the result. The |
4838 | /// address of the memory location does not have to be aligned. |
4839 | /// \returns A 256-bit integer vector containing the concatenated result. |
4840 | static __inline __m256i __DEFAULT_FN_ATTRS |
4841 | _mm256_loadu2_m128i(__m128i_u const *__addr_hi, __m128i_u const *__addr_lo) |
4842 | { |
4843 | __m256i __v256 = _mm256_castsi128_si256(_mm_loadu_si128(__addr_lo)); |
4844 | return _mm256_insertf128_si256(__v256, _mm_loadu_si128(__addr_hi), 1); |
4845 | } |
4846 | |
4847 | /* SIMD store ops (unaligned) */ |
4848 | /// Stores the upper and lower 128 bits of a 256-bit floating-point |
4849 | /// vector of [8 x float] into two different unaligned memory locations. |
4850 | /// |
4851 | /// \headerfile <x86intrin.h> |
4852 | /// |
4853 | /// This intrinsic corresponds to the <c> VEXTRACTF128 </c> instruction and the |
4854 | /// store instructions. |
4855 | /// |
4856 | /// \param __addr_hi |
4857 | /// A pointer to a 128-bit memory location. Bits[255:128] of \a __a are to be |
4858 | /// copied to this memory location. The address of this memory location does |
4859 | /// not have to be aligned. |
4860 | /// \param __addr_lo |
4861 | /// A pointer to a 128-bit memory location. Bits[127:0] of \a __a are to be |
4862 | /// copied to this memory location. The address of this memory location does |
4863 | /// not have to be aligned. |
4864 | /// \param __a |
4865 | /// A 256-bit floating-point vector of [8 x float]. |
4866 | static __inline void __DEFAULT_FN_ATTRS |
4867 | _mm256_storeu2_m128(float *__addr_hi, float *__addr_lo, __m256 __a) |
4868 | { |
4869 | __m128 __v128; |
4870 | |
4871 | __v128 = _mm256_castps256_ps128(__a); |
4872 | _mm_storeu_ps(__addr_lo, __v128); |
4873 | __v128 = _mm256_extractf128_ps(__a, 1); |
4874 | _mm_storeu_ps(__addr_hi, __v128); |
4875 | } |
4876 | |
4877 | /// Stores the upper and lower 128 bits of a 256-bit floating-point |
4878 | /// vector of [4 x double] into two different unaligned memory locations. |
4879 | /// |
4880 | /// \headerfile <x86intrin.h> |
4881 | /// |
4882 | /// This intrinsic corresponds to the <c> VEXTRACTF128 </c> instruction and the |
4883 | /// store instructions. |
4884 | /// |
4885 | /// \param __addr_hi |
4886 | /// A pointer to a 128-bit memory location. Bits[255:128] of \a __a are to be |
4887 | /// copied to this memory location. The address of this memory location does |
4888 | /// not have to be aligned. |
4889 | /// \param __addr_lo |
4890 | /// A pointer to a 128-bit memory location. Bits[127:0] of \a __a are to be |
4891 | /// copied to this memory location. The address of this memory location does |
4892 | /// not have to be aligned. |
4893 | /// \param __a |
4894 | /// A 256-bit floating-point vector of [4 x double]. |
4895 | static __inline void __DEFAULT_FN_ATTRS |
4896 | _mm256_storeu2_m128d(double *__addr_hi, double *__addr_lo, __m256d __a) |
4897 | { |
4898 | __m128d __v128; |
4899 | |
4900 | __v128 = _mm256_castpd256_pd128(__a); |
4901 | _mm_storeu_pd(__addr_lo, __v128); |
4902 | __v128 = _mm256_extractf128_pd(__a, 1); |
4903 | _mm_storeu_pd(__addr_hi, __v128); |
4904 | } |
4905 | |
4906 | /// Stores the upper and lower 128 bits of a 256-bit integer vector into |
4907 | /// two different unaligned memory locations. |
4908 | /// |
4909 | /// \headerfile <x86intrin.h> |
4910 | /// |
4911 | /// This intrinsic corresponds to the <c> VEXTRACTF128 </c> instruction and the |
4912 | /// store instructions. |
4913 | /// |
4914 | /// \param __addr_hi |
4915 | /// A pointer to a 128-bit memory location. Bits[255:128] of \a __a are to be |
4916 | /// copied to this memory location. The address of this memory location does |
4917 | /// not have to be aligned. |
4918 | /// \param __addr_lo |
4919 | /// A pointer to a 128-bit memory location. Bits[127:0] of \a __a are to be |
4920 | /// copied to this memory location. The address of this memory location does |
4921 | /// not have to be aligned. |
4922 | /// \param __a |
4923 | /// A 256-bit integer vector. |
4924 | static __inline void __DEFAULT_FN_ATTRS |
4925 | _mm256_storeu2_m128i(__m128i_u *__addr_hi, __m128i_u *__addr_lo, __m256i __a) |
4926 | { |
4927 | __m128i __v128; |
4928 | |
4929 | __v128 = _mm256_castsi256_si128(__a); |
4930 | _mm_storeu_si128(__addr_lo, __v128); |
4931 | __v128 = _mm256_extractf128_si256(__a, 1); |
4932 | _mm_storeu_si128(__addr_hi, __v128); |
4933 | } |
4934 | |
4935 | /// Constructs a 256-bit floating-point vector of [8 x float] by |
4936 | /// concatenating two 128-bit floating-point vectors of [4 x float]. |
4937 | /// |
4938 | /// \headerfile <x86intrin.h> |
4939 | /// |
4940 | /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction. |
4941 | /// |
4942 | /// \param __hi |
4943 | /// A 128-bit floating-point vector of [4 x float] to be copied to the upper |
4944 | /// 128 bits of the result. |
4945 | /// \param __lo |
4946 | /// A 128-bit floating-point vector of [4 x float] to be copied to the lower |
4947 | /// 128 bits of the result. |
4948 | /// \returns A 256-bit floating-point vector of [8 x float] containing the |
4949 | /// concatenated result. |
4950 | static __inline __m256 __DEFAULT_FN_ATTRS |
4951 | _mm256_set_m128 (__m128 __hi, __m128 __lo) |
4952 | { |
4953 | return (__m256) __builtin_shufflevector((__v4sf)__lo, (__v4sf)__hi, 0, 1, 2, 3, 4, 5, 6, 7); |
4954 | } |
4955 | |
4956 | /// Constructs a 256-bit floating-point vector of [4 x double] by |
4957 | /// concatenating two 128-bit floating-point vectors of [2 x double]. |
4958 | /// |
4959 | /// \headerfile <x86intrin.h> |
4960 | /// |
4961 | /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction. |
4962 | /// |
4963 | /// \param __hi |
4964 | /// A 128-bit floating-point vector of [2 x double] to be copied to the upper |
4965 | /// 128 bits of the result. |
4966 | /// \param __lo |
4967 | /// A 128-bit floating-point vector of [2 x double] to be copied to the lower |
4968 | /// 128 bits of the result. |
4969 | /// \returns A 256-bit floating-point vector of [4 x double] containing the |
4970 | /// concatenated result. |
4971 | static __inline __m256d __DEFAULT_FN_ATTRS |
4972 | _mm256_set_m128d (__m128d __hi, __m128d __lo) |
4973 | { |
4974 | return (__m256d) __builtin_shufflevector((__v2df)__lo, (__v2df)__hi, 0, 1, 2, 3); |
4975 | } |
4976 | |
4977 | /// Constructs a 256-bit integer vector by concatenating two 128-bit |
4978 | /// integer vectors. |
4979 | /// |
4980 | /// \headerfile <x86intrin.h> |
4981 | /// |
4982 | /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction. |
4983 | /// |
4984 | /// \param __hi |
4985 | /// A 128-bit integer vector to be copied to the upper 128 bits of the |
4986 | /// result. |
4987 | /// \param __lo |
4988 | /// A 128-bit integer vector to be copied to the lower 128 bits of the |
4989 | /// result. |
4990 | /// \returns A 256-bit integer vector containing the concatenated result. |
4991 | static __inline __m256i __DEFAULT_FN_ATTRS |
4992 | _mm256_set_m128i (__m128i __hi, __m128i __lo) |
4993 | { |
4994 | return (__m256i) __builtin_shufflevector((__v2di)__lo, (__v2di)__hi, 0, 1, 2, 3); |
4995 | } |
4996 | |
4997 | /// Constructs a 256-bit floating-point vector of [8 x float] by |
4998 | /// concatenating two 128-bit floating-point vectors of [4 x float]. This is |
4999 | /// similar to _mm256_set_m128, but the order of the input parameters is |
5000 | /// swapped. |
5001 | /// |
5002 | /// \headerfile <x86intrin.h> |
5003 | /// |
5004 | /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction. |
5005 | /// |
5006 | /// \param __lo |
5007 | /// A 128-bit floating-point vector of [4 x float] to be copied to the lower |
5008 | /// 128 bits of the result. |
5009 | /// \param __hi |
5010 | /// A 128-bit floating-point vector of [4 x float] to be copied to the upper |
5011 | /// 128 bits of the result. |
5012 | /// \returns A 256-bit floating-point vector of [8 x float] containing the |
5013 | /// concatenated result. |
5014 | static __inline __m256 __DEFAULT_FN_ATTRS |
5015 | _mm256_setr_m128 (__m128 __lo, __m128 __hi) |
5016 | { |
5017 | return _mm256_set_m128(__hi, __lo); |
5018 | } |
5019 | |
5020 | /// Constructs a 256-bit floating-point vector of [4 x double] by |
5021 | /// concatenating two 128-bit floating-point vectors of [2 x double]. This is |
5022 | /// similar to _mm256_set_m128d, but the order of the input parameters is |
5023 | /// swapped. |
5024 | /// |
5025 | /// \headerfile <x86intrin.h> |
5026 | /// |
5027 | /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction. |
5028 | /// |
5029 | /// \param __lo |
5030 | /// A 128-bit floating-point vector of [2 x double] to be copied to the lower |
5031 | /// 128 bits of the result. |
5032 | /// \param __hi |
5033 | /// A 128-bit floating-point vector of [2 x double] to be copied to the upper |
5034 | /// 128 bits of the result. |
5035 | /// \returns A 256-bit floating-point vector of [4 x double] containing the |
5036 | /// concatenated result. |
5037 | static __inline __m256d __DEFAULT_FN_ATTRS |
5038 | _mm256_setr_m128d (__m128d __lo, __m128d __hi) |
5039 | { |
5040 | return (__m256d)_mm256_set_m128d(__hi, __lo); |
5041 | } |
5042 | |
5043 | /// Constructs a 256-bit integer vector by concatenating two 128-bit |
5044 | /// integer vectors. This is similar to _mm256_set_m128i, but the order of |
5045 | /// the input parameters is swapped. |
5046 | /// |
5047 | /// \headerfile <x86intrin.h> |
5048 | /// |
5049 | /// This intrinsic corresponds to the <c> VINSERTF128 </c> instruction. |
5050 | /// |
5051 | /// \param __lo |
5052 | /// A 128-bit integer vector to be copied to the lower 128 bits of the |
5053 | /// result. |
5054 | /// \param __hi |
5055 | /// A 128-bit integer vector to be copied to the upper 128 bits of the |
5056 | /// result. |
5057 | /// \returns A 256-bit integer vector containing the concatenated result. |
5058 | static __inline __m256i __DEFAULT_FN_ATTRS |
5059 | _mm256_setr_m128i (__m128i __lo, __m128i __hi) |
5060 | { |
5061 | return (__m256i)_mm256_set_m128i(__hi, __lo); |
5062 | } |
5063 | |
5064 | #undef __DEFAULT_FN_ATTRS |
5065 | #undef __DEFAULT_FN_ATTRS128 |
5066 | |
5067 | #endif /* __AVXINTRIN_H */ |
5068 |