1 /*===------------- avx512bwintrin.h - AVX512BW intrinsics ------------------===
4 * Permission is hereby granted, free of charge, to any person obtaining a copy
5 * of this software and associated documentation files (the "Software"), to deal
6 * in the Software without restriction, including without limitation the rights
7 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
8 * copies of the Software, and to permit persons to whom the Software is
9 * furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
17 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
19 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 *===-----------------------------------------------------------------------===
25 #error "Never use <avx512bwintrin.h> directly; include <immintrin.h> instead."
28 #ifndef __AVX512BWINTRIN_H
29 #define __AVX512BWINTRIN_H
31 typedef unsigned int __mmask32
;
32 typedef unsigned long long __mmask64
;
33 typedef char __v64qi
__attribute__ ((__vector_size__ (64)));
34 typedef short __v32hi
__attribute__ ((__vector_size__ (64)));
36 /* Define the default attributes for the functions in this file. */
37 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512bw")))
39 static __inline __v64qi __DEFAULT_FN_ATTRS
40 _mm512_setzero_qi(void) {
41 return (__v64qi
){ 0, 0, 0, 0, 0, 0, 0, 0,
42 0, 0, 0, 0, 0, 0, 0, 0,
43 0, 0, 0, 0, 0, 0, 0, 0,
44 0, 0, 0, 0, 0, 0, 0, 0,
45 0, 0, 0, 0, 0, 0, 0, 0,
46 0, 0, 0, 0, 0, 0, 0, 0,
47 0, 0, 0, 0, 0, 0, 0, 0,
48 0, 0, 0, 0, 0, 0, 0, 0 };
51 static __inline __v32hi __DEFAULT_FN_ATTRS
52 _mm512_setzero_hi(void) {
53 return (__v32hi
){ 0, 0, 0, 0, 0, 0, 0, 0,
54 0, 0, 0, 0, 0, 0, 0, 0,
55 0, 0, 0, 0, 0, 0, 0, 0,
56 0, 0, 0, 0, 0, 0, 0, 0 };
61 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
62 _mm512_cmpeq_epi8_mask(__m512i __a
, __m512i __b
) {
63 return (__mmask64
)__builtin_ia32_pcmpeqb512_mask((__v64qi
)__a
, (__v64qi
)__b
,
67 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
68 _mm512_mask_cmpeq_epi8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
69 return (__mmask64
)__builtin_ia32_pcmpeqb512_mask((__v64qi
)__a
, (__v64qi
)__b
,
73 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
74 _mm512_cmpeq_epu8_mask(__m512i __a
, __m512i __b
) {
75 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 0,
79 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
80 _mm512_mask_cmpeq_epu8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
81 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 0,
85 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
86 _mm512_cmpeq_epi16_mask(__m512i __a
, __m512i __b
) {
87 return (__mmask32
)__builtin_ia32_pcmpeqw512_mask((__v32hi
)__a
, (__v32hi
)__b
,
91 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
92 _mm512_mask_cmpeq_epi16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
93 return (__mmask32
)__builtin_ia32_pcmpeqw512_mask((__v32hi
)__a
, (__v32hi
)__b
,
97 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
98 _mm512_cmpeq_epu16_mask(__m512i __a
, __m512i __b
) {
99 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 0,
103 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
104 _mm512_mask_cmpeq_epu16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
105 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 0,
109 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
110 _mm512_cmpge_epi8_mask(__m512i __a
, __m512i __b
) {
111 return (__mmask64
)__builtin_ia32_cmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 5,
115 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
116 _mm512_mask_cmpge_epi8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
117 return (__mmask64
)__builtin_ia32_cmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 5,
121 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
122 _mm512_cmpge_epu8_mask(__m512i __a
, __m512i __b
) {
123 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 5,
127 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
128 _mm512_mask_cmpge_epu8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
129 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 5,
133 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
134 _mm512_cmpge_epi16_mask(__m512i __a
, __m512i __b
) {
135 return (__mmask32
)__builtin_ia32_cmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 5,
139 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
140 _mm512_mask_cmpge_epi16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
141 return (__mmask32
)__builtin_ia32_cmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 5,
145 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
146 _mm512_cmpge_epu16_mask(__m512i __a
, __m512i __b
) {
147 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 5,
151 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
152 _mm512_mask_cmpge_epu16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
153 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 5,
157 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
158 _mm512_cmpgt_epi8_mask(__m512i __a
, __m512i __b
) {
159 return (__mmask64
)__builtin_ia32_pcmpgtb512_mask((__v64qi
)__a
, (__v64qi
)__b
,
163 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
164 _mm512_mask_cmpgt_epi8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
165 return (__mmask64
)__builtin_ia32_pcmpgtb512_mask((__v64qi
)__a
, (__v64qi
)__b
,
169 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
170 _mm512_cmpgt_epu8_mask(__m512i __a
, __m512i __b
) {
171 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 6,
175 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
176 _mm512_mask_cmpgt_epu8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
177 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 6,
181 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
182 _mm512_cmpgt_epi16_mask(__m512i __a
, __m512i __b
) {
183 return (__mmask32
)__builtin_ia32_pcmpgtw512_mask((__v32hi
)__a
, (__v32hi
)__b
,
187 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
188 _mm512_mask_cmpgt_epi16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
189 return (__mmask32
)__builtin_ia32_pcmpgtw512_mask((__v32hi
)__a
, (__v32hi
)__b
,
193 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
194 _mm512_cmpgt_epu16_mask(__m512i __a
, __m512i __b
) {
195 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 6,
199 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
200 _mm512_mask_cmpgt_epu16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
201 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 6,
205 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
206 _mm512_cmple_epi8_mask(__m512i __a
, __m512i __b
) {
207 return (__mmask64
)__builtin_ia32_cmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 2,
211 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
212 _mm512_mask_cmple_epi8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
213 return (__mmask64
)__builtin_ia32_cmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 2,
217 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
218 _mm512_cmple_epu8_mask(__m512i __a
, __m512i __b
) {
219 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 2,
223 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
224 _mm512_mask_cmple_epu8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
225 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 2,
229 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
230 _mm512_cmple_epi16_mask(__m512i __a
, __m512i __b
) {
231 return (__mmask32
)__builtin_ia32_cmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 2,
235 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
236 _mm512_mask_cmple_epi16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
237 return (__mmask32
)__builtin_ia32_cmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 2,
241 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
242 _mm512_cmple_epu16_mask(__m512i __a
, __m512i __b
) {
243 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 2,
247 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
248 _mm512_mask_cmple_epu16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
249 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 2,
253 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
254 _mm512_cmplt_epi8_mask(__m512i __a
, __m512i __b
) {
255 return (__mmask64
)__builtin_ia32_cmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 1,
259 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
260 _mm512_mask_cmplt_epi8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
261 return (__mmask64
)__builtin_ia32_cmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 1,
265 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
266 _mm512_cmplt_epu8_mask(__m512i __a
, __m512i __b
) {
267 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 1,
271 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
272 _mm512_mask_cmplt_epu8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
273 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 1,
277 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
278 _mm512_cmplt_epi16_mask(__m512i __a
, __m512i __b
) {
279 return (__mmask32
)__builtin_ia32_cmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 1,
283 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
284 _mm512_mask_cmplt_epi16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
285 return (__mmask32
)__builtin_ia32_cmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 1,
289 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
290 _mm512_cmplt_epu16_mask(__m512i __a
, __m512i __b
) {
291 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 1,
295 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
296 _mm512_mask_cmplt_epu16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
297 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 1,
301 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
302 _mm512_cmpneq_epi8_mask(__m512i __a
, __m512i __b
) {
303 return (__mmask64
)__builtin_ia32_cmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 4,
307 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
308 _mm512_mask_cmpneq_epi8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
309 return (__mmask64
)__builtin_ia32_cmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 4,
313 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
314 _mm512_cmpneq_epu8_mask(__m512i __a
, __m512i __b
) {
315 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 4,
319 static __inline__ __mmask64 __DEFAULT_FN_ATTRS
320 _mm512_mask_cmpneq_epu8_mask(__mmask64 __u
, __m512i __a
, __m512i __b
) {
321 return (__mmask64
)__builtin_ia32_ucmpb512_mask((__v64qi
)__a
, (__v64qi
)__b
, 4,
325 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
326 _mm512_cmpneq_epi16_mask(__m512i __a
, __m512i __b
) {
327 return (__mmask32
)__builtin_ia32_cmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 4,
331 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
332 _mm512_mask_cmpneq_epi16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
333 return (__mmask32
)__builtin_ia32_cmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 4,
337 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
338 _mm512_cmpneq_epu16_mask(__m512i __a
, __m512i __b
) {
339 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 4,
343 static __inline__ __mmask32 __DEFAULT_FN_ATTRS
344 _mm512_mask_cmpneq_epu16_mask(__mmask32 __u
, __m512i __a
, __m512i __b
) {
345 return (__mmask32
)__builtin_ia32_ucmpw512_mask((__v32hi
)__a
, (__v32hi
)__b
, 4,
349 static __inline__ __m512i __DEFAULT_FN_ATTRS
350 _mm512_add_epi8 (__m512i __A
, __m512i __B
) {
351 return (__m512i
) ((__v64qi
) __A
+ (__v64qi
) __B
);
354 static __inline__ __m512i __DEFAULT_FN_ATTRS
355 _mm512_mask_add_epi8 (__m512i __W
, __mmask64 __U
, __m512i __A
, __m512i __B
) {
356 return (__m512i
) __builtin_ia32_paddb512_mask ((__v64qi
) __A
,
362 static __inline__ __m512i __DEFAULT_FN_ATTRS
363 _mm512_maskz_add_epi8 (__mmask64 __U
, __m512i __A
, __m512i __B
) {
364 return (__m512i
) __builtin_ia32_paddb512_mask ((__v64qi
) __A
,
366 (__v64qi
) _mm512_setzero_qi(),
370 static __inline__ __m512i __DEFAULT_FN_ATTRS
371 _mm512_sub_epi8 (__m512i __A
, __m512i __B
) {
372 return (__m512i
) ((__v64qi
) __A
- (__v64qi
) __B
);
375 static __inline__ __m512i __DEFAULT_FN_ATTRS
376 _mm512_mask_sub_epi8 (__m512i __W
, __mmask64 __U
, __m512i __A
, __m512i __B
) {
377 return (__m512i
) __builtin_ia32_psubb512_mask ((__v64qi
) __A
,
383 static __inline__ __m512i __DEFAULT_FN_ATTRS
384 _mm512_maskz_sub_epi8 (__mmask64 __U
, __m512i __A
, __m512i __B
) {
385 return (__m512i
) __builtin_ia32_psubb512_mask ((__v64qi
) __A
,
387 (__v64qi
) _mm512_setzero_qi(),
391 static __inline__ __m512i __DEFAULT_FN_ATTRS
392 _mm512_add_epi16 (__m512i __A
, __m512i __B
) {
393 return (__m512i
) ((__v32hi
) __A
+ (__v32hi
) __B
);
396 static __inline__ __m512i __DEFAULT_FN_ATTRS
397 _mm512_mask_add_epi16 (__m512i __W
, __mmask32 __U
, __m512i __A
, __m512i __B
) {
398 return (__m512i
) __builtin_ia32_paddw512_mask ((__v32hi
) __A
,
404 static __inline__ __m512i __DEFAULT_FN_ATTRS
405 _mm512_maskz_add_epi16 (__mmask32 __U
, __m512i __A
, __m512i __B
) {
406 return (__m512i
) __builtin_ia32_paddw512_mask ((__v32hi
) __A
,
408 (__v32hi
) _mm512_setzero_hi(),
412 static __inline__ __m512i __DEFAULT_FN_ATTRS
413 _mm512_sub_epi16 (__m512i __A
, __m512i __B
) {
414 return (__m512i
) ((__v32hi
) __A
- (__v32hi
) __B
);
417 static __inline__ __m512i __DEFAULT_FN_ATTRS
418 _mm512_mask_sub_epi16 (__m512i __W
, __mmask32 __U
, __m512i __A
, __m512i __B
) {
419 return (__m512i
) __builtin_ia32_psubw512_mask ((__v32hi
) __A
,
425 static __inline__ __m512i __DEFAULT_FN_ATTRS
426 _mm512_maskz_sub_epi16 (__mmask32 __U
, __m512i __A
, __m512i __B
) {
427 return (__m512i
) __builtin_ia32_psubw512_mask ((__v32hi
) __A
,
429 (__v32hi
) _mm512_setzero_hi(),
433 static __inline__ __m512i __DEFAULT_FN_ATTRS
434 _mm512_mullo_epi16 (__m512i __A
, __m512i __B
) {
435 return (__m512i
) ((__v32hi
) __A
* (__v32hi
) __B
);
438 static __inline__ __m512i __DEFAULT_FN_ATTRS
439 _mm512_mask_mullo_epi16 (__m512i __W
, __mmask32 __U
, __m512i __A
, __m512i __B
) {
440 return (__m512i
) __builtin_ia32_pmullw512_mask ((__v32hi
) __A
,
446 static __inline__ __m512i __DEFAULT_FN_ATTRS
447 _mm512_maskz_mullo_epi16 (__mmask32 __U
, __m512i __A
, __m512i __B
) {
448 return (__m512i
) __builtin_ia32_pmullw512_mask ((__v32hi
) __A
,
450 (__v32hi
) _mm512_setzero_hi(),
454 static __inline__ __m512i __DEFAULT_FN_ATTRS
455 _mm512_mask_blend_epi8 (__mmask64 __U
, __m512i __A
, __m512i __W
)
457 return (__m512i
) __builtin_ia32_blendmb_512_mask ((__v64qi
) __A
,
462 static __inline__ __m512i __DEFAULT_FN_ATTRS
463 _mm512_mask_blend_epi16 (__mmask32 __U
, __m512i __A
, __m512i __W
)
465 return (__m512i
) __builtin_ia32_blendmw_512_mask ((__v32hi
) __A
,
470 static __inline__ __m512i __DEFAULT_FN_ATTRS
471 _mm512_abs_epi8 (__m512i __A
)
473 return (__m512i
) __builtin_ia32_pabsb512_mask ((__v64qi
) __A
,
474 (__v64qi
) _mm512_setzero_qi(),
478 static __inline__ __m512i __DEFAULT_FN_ATTRS
479 _mm512_mask_abs_epi8 (__m512i __W
, __mmask64 __U
, __m512i __A
)
481 return (__m512i
) __builtin_ia32_pabsb512_mask ((__v64qi
) __A
,
486 static __inline__ __m512i __DEFAULT_FN_ATTRS
487 _mm512_maskz_abs_epi8 (__mmask64 __U
, __m512i __A
)
489 return (__m512i
) __builtin_ia32_pabsb512_mask ((__v64qi
) __A
,
490 (__v64qi
) _mm512_setzero_qi(),
494 static __inline__ __m512i __DEFAULT_FN_ATTRS
495 _mm512_abs_epi16 (__m512i __A
)
497 return (__m512i
) __builtin_ia32_pabsw512_mask ((__v32hi
) __A
,
498 (__v32hi
) _mm512_setzero_hi(),
502 static __inline__ __m512i __DEFAULT_FN_ATTRS
503 _mm512_mask_abs_epi16 (__m512i __W
, __mmask32 __U
, __m512i __A
)
505 return (__m512i
) __builtin_ia32_pabsw512_mask ((__v32hi
) __A
,
510 static __inline__ __m512i __DEFAULT_FN_ATTRS
511 _mm512_maskz_abs_epi16 (__mmask32 __U
, __m512i __A
)
513 return (__m512i
) __builtin_ia32_pabsw512_mask ((__v32hi
) __A
,
514 (__v32hi
) _mm512_setzero_hi(),
518 static __inline__ __m512i __DEFAULT_FN_ATTRS
519 _mm512_packs_epi32 (__m512i __A
, __m512i __B
)
521 return (__m512i
) __builtin_ia32_packssdw512_mask ((__v16si
) __A
,
523 (__v32hi
) _mm512_setzero_hi(),
527 static __inline__ __m512i __DEFAULT_FN_ATTRS
528 _mm512_maskz_packs_epi32 (__mmask32 __M
, __m512i __A
, __m512i __B
)
530 return (__m512i
) __builtin_ia32_packssdw512_mask ((__v16si
) __A
,
532 (__v32hi
) _mm512_setzero_hi(),
536 static __inline__ __m512i __DEFAULT_FN_ATTRS
537 _mm512_mask_packs_epi32 (__m512i __W
, __mmask32 __M
, __m512i __A
,
540 return (__m512i
) __builtin_ia32_packssdw512_mask ((__v16si
) __A
,
546 static __inline__ __m512i __DEFAULT_FN_ATTRS
547 _mm512_packs_epi16 (__m512i __A
, __m512i __B
)
549 return (__m512i
) __builtin_ia32_packsswb512_mask ((__v32hi
) __A
,
551 (__v64qi
) _mm512_setzero_qi(),
555 static __inline__ __m512i __DEFAULT_FN_ATTRS
556 _mm512_mask_packs_epi16 (__m512i __W
, __mmask64 __M
, __m512i __A
,
559 return (__m512i
) __builtin_ia32_packsswb512_mask ((__v32hi
) __A
,
565 static __inline__ __m512i __DEFAULT_FN_ATTRS
566 _mm512_maskz_packs_epi16 (__mmask64 __M
, __m512i __A
, __m512i __B
)
568 return (__m512i
) __builtin_ia32_packsswb512_mask ((__v32hi
) __A
,
570 (__v64qi
) _mm512_setzero_qi(),
574 static __inline__ __m512i __DEFAULT_FN_ATTRS
575 _mm512_packus_epi32 (__m512i __A
, __m512i __B
)
577 return (__m512i
) __builtin_ia32_packusdw512_mask ((__v16si
) __A
,
579 (__v32hi
) _mm512_setzero_hi(),
583 static __inline__ __m512i __DEFAULT_FN_ATTRS
584 _mm512_maskz_packus_epi32 (__mmask32 __M
, __m512i __A
, __m512i __B
)
586 return (__m512i
) __builtin_ia32_packusdw512_mask ((__v16si
) __A
,
588 (__v32hi
) _mm512_setzero_hi(),
592 static __inline__ __m512i __DEFAULT_FN_ATTRS
593 _mm512_mask_packus_epi32 (__m512i __W
, __mmask32 __M
, __m512i __A
,
596 return (__m512i
) __builtin_ia32_packusdw512_mask ((__v16si
) __A
,
602 static __inline__ __m512i __DEFAULT_FN_ATTRS
603 _mm512_packus_epi16 (__m512i __A
, __m512i __B
)
605 return (__m512i
) __builtin_ia32_packuswb512_mask ((__v32hi
) __A
,
607 (__v64qi
) _mm512_setzero_qi(),
611 static __inline__ __m512i __DEFAULT_FN_ATTRS
612 _mm512_mask_packus_epi16 (__m512i __W
, __mmask64 __M
, __m512i __A
,
615 return (__m512i
) __builtin_ia32_packuswb512_mask ((__v32hi
) __A
,
621 static __inline__ __m512i __DEFAULT_FN_ATTRS
622 _mm512_maskz_packus_epi16 (__mmask64 __M
, __m512i __A
, __m512i __B
)
624 return (__m512i
) __builtin_ia32_packuswb512_mask ((__v32hi
) __A
,
626 (__v64qi
) _mm512_setzero_qi(),
630 static __inline__ __m512i __DEFAULT_FN_ATTRS
631 _mm512_adds_epi8 (__m512i __A
, __m512i __B
)
633 return (__m512i
) __builtin_ia32_paddsb512_mask ((__v64qi
) __A
,
635 (__v64qi
) _mm512_setzero_qi(),
639 static __inline__ __m512i __DEFAULT_FN_ATTRS
640 _mm512_mask_adds_epi8 (__m512i __W
, __mmask64 __U
, __m512i __A
,
643 return (__m512i
) __builtin_ia32_paddsb512_mask ((__v64qi
) __A
,
649 static __inline__ __m512i __DEFAULT_FN_ATTRS
650 _mm512_maskz_adds_epi8 (__mmask64 __U
, __m512i __A
, __m512i __B
)
652 return (__m512i
) __builtin_ia32_paddsb512_mask ((__v64qi
) __A
,
654 (__v64qi
) _mm512_setzero_qi(),
658 static __inline__ __m512i __DEFAULT_FN_ATTRS
659 _mm512_adds_epi16 (__m512i __A
, __m512i __B
)
661 return (__m512i
) __builtin_ia32_paddsw512_mask ((__v32hi
) __A
,
663 (__v32hi
) _mm512_setzero_hi(),
667 static __inline__ __m512i __DEFAULT_FN_ATTRS
668 _mm512_mask_adds_epi16 (__m512i __W
, __mmask32 __U
, __m512i __A
,
671 return (__m512i
) __builtin_ia32_paddsw512_mask ((__v32hi
) __A
,
677 static __inline__ __m512i __DEFAULT_FN_ATTRS
678 _mm512_maskz_adds_epi16 (__mmask32 __U
, __m512i __A
, __m512i __B
)
680 return (__m512i
) __builtin_ia32_paddsw512_mask ((__v32hi
) __A
,
682 (__v32hi
) _mm512_setzero_hi(),
686 static __inline__ __m512i __DEFAULT_FN_ATTRS
687 _mm512_adds_epu8 (__m512i __A
, __m512i __B
)
689 return (__m512i
) __builtin_ia32_paddusb512_mask ((__v64qi
) __A
,
691 (__v64qi
) _mm512_setzero_qi(),
695 static __inline__ __m512i __DEFAULT_FN_ATTRS
696 _mm512_mask_adds_epu8 (__m512i __W
, __mmask64 __U
, __m512i __A
,
699 return (__m512i
) __builtin_ia32_paddusb512_mask ((__v64qi
) __A
,
705 static __inline__ __m512i __DEFAULT_FN_ATTRS
706 _mm512_maskz_adds_epu8 (__mmask64 __U
, __m512i __A
, __m512i __B
)
708 return (__m512i
) __builtin_ia32_paddusb512_mask ((__v64qi
) __A
,
710 (__v64qi
) _mm512_setzero_qi(),
714 static __inline__ __m512i __DEFAULT_FN_ATTRS
715 _mm512_adds_epu16 (__m512i __A
, __m512i __B
)
717 return (__m512i
) __builtin_ia32_paddusw512_mask ((__v32hi
) __A
,
719 (__v32hi
) _mm512_setzero_hi(),
723 static __inline__ __m512i __DEFAULT_FN_ATTRS
724 _mm512_mask_adds_epu16 (__m512i __W
, __mmask32 __U
, __m512i __A
,
727 return (__m512i
) __builtin_ia32_paddusw512_mask ((__v32hi
) __A
,
733 static __inline__ __m512i __DEFAULT_FN_ATTRS
734 _mm512_maskz_adds_epu16 (__mmask32 __U
, __m512i __A
, __m512i __B
)
736 return (__m512i
) __builtin_ia32_paddusw512_mask ((__v32hi
) __A
,
738 (__v32hi
) _mm512_setzero_hi(),
742 static __inline__ __m512i __DEFAULT_FN_ATTRS
743 _mm512_avg_epu8 (__m512i __A
, __m512i __B
)
745 return (__m512i
) __builtin_ia32_pavgb512_mask ((__v64qi
) __A
,
747 (__v64qi
) _mm512_setzero_qi(),
751 static __inline__ __m512i __DEFAULT_FN_ATTRS
752 _mm512_mask_avg_epu8 (__m512i __W
, __mmask64 __U
, __m512i __A
,
755 return (__m512i
) __builtin_ia32_pavgb512_mask ((__v64qi
) __A
,
761 static __inline__ __m512i __DEFAULT_FN_ATTRS
762 _mm512_maskz_avg_epu8 (__mmask64 __U
, __m512i __A
, __m512i __B
)
764 return (__m512i
) __builtin_ia32_pavgb512_mask ((__v64qi
) __A
,
766 (__v64qi
) _mm512_setzero_qi(),
770 static __inline__ __m512i __DEFAULT_FN_ATTRS
771 _mm512_avg_epu16 (__m512i __A
, __m512i __B
)
773 return (__m512i
) __builtin_ia32_pavgw512_mask ((__v32hi
) __A
,
775 (__v32hi
) _mm512_setzero_hi(),
779 static __inline__ __m512i __DEFAULT_FN_ATTRS
780 _mm512_mask_avg_epu16 (__m512i __W
, __mmask32 __U
, __m512i __A
,
783 return (__m512i
) __builtin_ia32_pavgw512_mask ((__v32hi
) __A
,
789 static __inline__ __m512i __DEFAULT_FN_ATTRS
790 _mm512_maskz_avg_epu16 (__mmask32 __U
, __m512i __A
, __m512i __B
)
792 return (__m512i
) __builtin_ia32_pavgw512_mask ((__v32hi
) __A
,
794 (__v32hi
) _mm512_setzero_hi(),
798 static __inline__ __m512i __DEFAULT_FN_ATTRS
799 _mm512_max_epi8 (__m512i __A
, __m512i __B
)
801 return (__m512i
) __builtin_ia32_pmaxsb512_mask ((__v64qi
) __A
,
803 (__v64qi
) _mm512_setzero_qi(),
807 static __inline__ __m512i __DEFAULT_FN_ATTRS
808 _mm512_maskz_max_epi8 (__mmask64 __M
, __m512i __A
, __m512i __B
)
810 return (__m512i
) __builtin_ia32_pmaxsb512_mask ((__v64qi
) __A
,
812 (__v64qi
) _mm512_setzero_qi(),
816 static __inline__ __m512i __DEFAULT_FN_ATTRS
817 _mm512_mask_max_epi8 (__m512i __W
, __mmask64 __M
, __m512i __A
,
820 return (__m512i
) __builtin_ia32_pmaxsb512_mask ((__v64qi
) __A
,
826 static __inline__ __m512i __DEFAULT_FN_ATTRS
827 _mm512_max_epi16 (__m512i __A
, __m512i __B
)
829 return (__m512i
) __builtin_ia32_pmaxsw512_mask ((__v32hi
) __A
,
831 (__v32hi
) _mm512_setzero_hi(),
835 static __inline__ __m512i __DEFAULT_FN_ATTRS
836 _mm512_maskz_max_epi16 (__mmask32 __M
, __m512i __A
, __m512i __B
)
838 return (__m512i
) __builtin_ia32_pmaxsw512_mask ((__v32hi
) __A
,
840 (__v32hi
) _mm512_setzero_hi(),
844 static __inline__ __m512i __DEFAULT_FN_ATTRS
845 _mm512_mask_max_epi16 (__m512i __W
, __mmask32 __M
, __m512i __A
,
848 return (__m512i
) __builtin_ia32_pmaxsw512_mask ((__v32hi
) __A
,
854 static __inline__ __m512i __DEFAULT_FN_ATTRS
855 _mm512_max_epu8 (__m512i __A
, __m512i __B
)
857 return (__m512i
) __builtin_ia32_pmaxub512_mask ((__v64qi
) __A
,
859 (__v64qi
) _mm512_setzero_qi(),
863 static __inline__ __m512i __DEFAULT_FN_ATTRS
864 _mm512_maskz_max_epu8 (__mmask64 __M
, __m512i __A
, __m512i __B
)
866 return (__m512i
) __builtin_ia32_pmaxub512_mask ((__v64qi
) __A
,
868 (__v64qi
) _mm512_setzero_qi(),
872 static __inline__ __m512i __DEFAULT_FN_ATTRS
873 _mm512_mask_max_epu8 (__m512i __W
, __mmask64 __M
, __m512i __A
,
876 return (__m512i
) __builtin_ia32_pmaxub512_mask ((__v64qi
) __A
,
882 static __inline__ __m512i __DEFAULT_FN_ATTRS
883 _mm512_max_epu16 (__m512i __A
, __m512i __B
)
885 return (__m512i
) __builtin_ia32_pmaxuw512_mask ((__v32hi
) __A
,
887 (__v32hi
) _mm512_setzero_hi(),
891 static __inline__ __m512i __DEFAULT_FN_ATTRS
892 _mm512_maskz_max_epu16 (__mmask32 __M
, __m512i __A
, __m512i __B
)
894 return (__m512i
) __builtin_ia32_pmaxuw512_mask ((__v32hi
) __A
,
896 (__v32hi
) _mm512_setzero_hi(),
900 static __inline__ __m512i __DEFAULT_FN_ATTRS
901 _mm512_mask_max_epu16 (__m512i __W
, __mmask32 __M
, __m512i __A
,
904 return (__m512i
) __builtin_ia32_pmaxuw512_mask ((__v32hi
) __A
,
910 static __inline__ __m512i __DEFAULT_FN_ATTRS
911 _mm512_min_epi8 (__m512i __A
, __m512i __B
)
913 return (__m512i
) __builtin_ia32_pminsb512_mask ((__v64qi
) __A
,
915 (__v64qi
) _mm512_setzero_qi(),
919 static __inline__ __m512i __DEFAULT_FN_ATTRS
920 _mm512_maskz_min_epi8 (__mmask64 __M
, __m512i __A
, __m512i __B
)
922 return (__m512i
) __builtin_ia32_pminsb512_mask ((__v64qi
) __A
,
924 (__v64qi
) _mm512_setzero_qi(),
928 static __inline__ __m512i __DEFAULT_FN_ATTRS
929 _mm512_mask_min_epi8 (__m512i __W
, __mmask64 __M
, __m512i __A
,
932 return (__m512i
) __builtin_ia32_pminsb512_mask ((__v64qi
) __A
,
938 static __inline__ __m512i __DEFAULT_FN_ATTRS
939 _mm512_min_epi16 (__m512i __A
, __m512i __B
)
941 return (__m512i
) __builtin_ia32_pminsw512_mask ((__v32hi
) __A
,
943 (__v32hi
) _mm512_setzero_hi(),
947 static __inline__ __m512i __DEFAULT_FN_ATTRS
948 _mm512_maskz_min_epi16 (__mmask32 __M
, __m512i __A
, __m512i __B
)
950 return (__m512i
) __builtin_ia32_pminsw512_mask ((__v32hi
) __A
,
952 (__v32hi
) _mm512_setzero_hi(),
956 static __inline__ __m512i __DEFAULT_FN_ATTRS
957 _mm512_mask_min_epi16 (__m512i __W
, __mmask32 __M
, __m512i __A
,
960 return (__m512i
) __builtin_ia32_pminsw512_mask ((__v32hi
) __A
,
966 static __inline__ __m512i __DEFAULT_FN_ATTRS
967 _mm512_min_epu8 (__m512i __A
, __m512i __B
)
969 return (__m512i
) __builtin_ia32_pminub512_mask ((__v64qi
) __A
,
971 (__v64qi
) _mm512_setzero_qi(),
975 static __inline__ __m512i __DEFAULT_FN_ATTRS
976 _mm512_maskz_min_epu8 (__mmask64 __M
, __m512i __A
, __m512i __B
)
978 return (__m512i
) __builtin_ia32_pminub512_mask ((__v64qi
) __A
,
980 (__v64qi
) _mm512_setzero_qi(),
984 static __inline__ __m512i __DEFAULT_FN_ATTRS
985 _mm512_mask_min_epu8 (__m512i __W
, __mmask64 __M
, __m512i __A
,
988 return (__m512i
) __builtin_ia32_pminub512_mask ((__v64qi
) __A
,
994 static __inline__ __m512i __DEFAULT_FN_ATTRS
995 _mm512_min_epu16 (__m512i __A
, __m512i __B
)
997 return (__m512i
) __builtin_ia32_pminuw512_mask ((__v32hi
) __A
,
999 (__v32hi
) _mm512_setzero_hi(),
1003 static __inline__ __m512i __DEFAULT_FN_ATTRS
1004 _mm512_maskz_min_epu16 (__mmask32 __M
, __m512i __A
, __m512i __B
)
1006 return (__m512i
) __builtin_ia32_pminuw512_mask ((__v32hi
) __A
,
1008 (__v32hi
) _mm512_setzero_hi(),
1012 static __inline__ __m512i __DEFAULT_FN_ATTRS
1013 _mm512_mask_min_epu16 (__m512i __W
, __mmask32 __M
, __m512i __A
,
1016 return (__m512i
) __builtin_ia32_pminuw512_mask ((__v32hi
) __A
,
1022 static __inline__ __m512i __DEFAULT_FN_ATTRS
1023 _mm512_shuffle_epi8 (__m512i __A
, __m512i __B
)
1025 return (__m512i
) __builtin_ia32_pshufb512_mask ((__v64qi
) __A
,
1027 (__v64qi
) _mm512_setzero_qi(),
1031 static __inline__ __m512i __DEFAULT_FN_ATTRS
1032 _mm512_mask_shuffle_epi8 (__m512i __W
, __mmask64 __U
, __m512i __A
,
1035 return (__m512i
) __builtin_ia32_pshufb512_mask ((__v64qi
) __A
,
1041 static __inline__ __m512i __DEFAULT_FN_ATTRS
1042 _mm512_maskz_shuffle_epi8 (__mmask64 __U
, __m512i __A
, __m512i __B
)
1044 return (__m512i
) __builtin_ia32_pshufb512_mask ((__v64qi
) __A
,
1046 (__v64qi
) _mm512_setzero_qi(),
1050 static __inline__ __m512i __DEFAULT_FN_ATTRS
1051 _mm512_subs_epi8 (__m512i __A
, __m512i __B
)
1053 return (__m512i
) __builtin_ia32_psubsb512_mask ((__v64qi
) __A
,
1055 (__v64qi
) _mm512_setzero_qi(),
1059 static __inline__ __m512i __DEFAULT_FN_ATTRS
1060 _mm512_mask_subs_epi8 (__m512i __W
, __mmask64 __U
, __m512i __A
,
1063 return (__m512i
) __builtin_ia32_psubsb512_mask ((__v64qi
) __A
,
1069 static __inline__ __m512i __DEFAULT_FN_ATTRS
1070 _mm512_maskz_subs_epi8 (__mmask64 __U
, __m512i __A
, __m512i __B
)
1072 return (__m512i
) __builtin_ia32_psubsb512_mask ((__v64qi
) __A
,
1074 (__v64qi
) _mm512_setzero_qi(),
1078 static __inline__ __m512i __DEFAULT_FN_ATTRS
1079 _mm512_subs_epi16 (__m512i __A
, __m512i __B
)
1081 return (__m512i
) __builtin_ia32_psubsw512_mask ((__v32hi
) __A
,
1083 (__v32hi
) _mm512_setzero_hi(),
1087 static __inline__ __m512i __DEFAULT_FN_ATTRS
1088 _mm512_mask_subs_epi16 (__m512i __W
, __mmask32 __U
, __m512i __A
,
1091 return (__m512i
) __builtin_ia32_psubsw512_mask ((__v32hi
) __A
,
1097 static __inline__ __m512i __DEFAULT_FN_ATTRS
1098 _mm512_maskz_subs_epi16 (__mmask32 __U
, __m512i __A
, __m512i __B
)
1100 return (__m512i
) __builtin_ia32_psubsw512_mask ((__v32hi
) __A
,
1102 (__v32hi
) _mm512_setzero_hi(),
1106 static __inline__ __m512i __DEFAULT_FN_ATTRS
1107 _mm512_subs_epu8 (__m512i __A
, __m512i __B
)
1109 return (__m512i
) __builtin_ia32_psubusb512_mask ((__v64qi
) __A
,
1111 (__v64qi
) _mm512_setzero_qi(),
1115 static __inline__ __m512i __DEFAULT_FN_ATTRS
1116 _mm512_mask_subs_epu8 (__m512i __W
, __mmask64 __U
, __m512i __A
,
1119 return (__m512i
) __builtin_ia32_psubusb512_mask ((__v64qi
) __A
,
1125 static __inline__ __m512i __DEFAULT_FN_ATTRS
1126 _mm512_maskz_subs_epu8 (__mmask64 __U
, __m512i __A
, __m512i __B
)
1128 return (__m512i
) __builtin_ia32_psubusb512_mask ((__v64qi
) __A
,
1130 (__v64qi
) _mm512_setzero_qi(),
1134 static __inline__ __m512i __DEFAULT_FN_ATTRS
1135 _mm512_subs_epu16 (__m512i __A
, __m512i __B
)
1137 return (__m512i
) __builtin_ia32_psubusw512_mask ((__v32hi
) __A
,
1139 (__v32hi
) _mm512_setzero_hi(),
1143 static __inline__ __m512i __DEFAULT_FN_ATTRS
1144 _mm512_mask_subs_epu16 (__m512i __W
, __mmask32 __U
, __m512i __A
,
1147 return (__m512i
) __builtin_ia32_psubusw512_mask ((__v32hi
) __A
,
1153 static __inline__ __m512i __DEFAULT_FN_ATTRS
1154 _mm512_maskz_subs_epu16 (__mmask32 __U
, __m512i __A
, __m512i __B
)
1156 return (__m512i
) __builtin_ia32_psubusw512_mask ((__v32hi
) __A
,
1158 (__v32hi
) _mm512_setzero_hi(),
1162 static __inline__ __m512i __DEFAULT_FN_ATTRS
1163 _mm512_mask2_permutex2var_epi16 (__m512i __A
, __m512i __I
,
1164 __mmask32 __U
, __m512i __B
)
1166 return (__m512i
) __builtin_ia32_vpermi2varhi512_mask ((__v32hi
) __A
,
1167 (__v32hi
) __I
/* idx */ ,
1172 static __inline__ __m512i __DEFAULT_FN_ATTRS
1173 _mm512_permutex2var_epi16 (__m512i __A
, __m512i __I
, __m512i __B
)
1175 return (__m512i
) __builtin_ia32_vpermt2varhi512_mask ((__v32hi
) __I
/* idx */,
1181 static __inline__ __m512i __DEFAULT_FN_ATTRS
1182 _mm512_mask_permutex2var_epi16 (__m512i __A
, __mmask32 __U
,
1183 __m512i __I
, __m512i __B
)
1185 return (__m512i
) __builtin_ia32_vpermt2varhi512_mask ((__v32hi
) __I
/* idx */,
1191 static __inline__ __m512i __DEFAULT_FN_ATTRS
1192 _mm512_maskz_permutex2var_epi16 (__mmask32 __U
, __m512i __A
,
1193 __m512i __I
, __m512i __B
)
1195 return (__m512i
) __builtin_ia32_vpermt2varhi512_maskz ((__v32hi
) __I
1202 static __inline__ __m512i __DEFAULT_FN_ATTRS
1203 _mm512_mulhrs_epi16 (__m512i __A
, __m512i __B
)
1205 return (__m512i
) __builtin_ia32_pmulhrsw512_mask ((__v32hi
) __A
,
1207 (__v32hi
) _mm512_setzero_hi(),
1211 static __inline__ __m512i __DEFAULT_FN_ATTRS
1212 _mm512_mask_mulhrs_epi16 (__m512i __W
, __mmask32 __U
, __m512i __A
,
1215 return (__m512i
) __builtin_ia32_pmulhrsw512_mask ((__v32hi
) __A
,
1221 static __inline__ __m512i __DEFAULT_FN_ATTRS
1222 _mm512_maskz_mulhrs_epi16 (__mmask32 __U
, __m512i __A
, __m512i __B
)
1224 return (__m512i
) __builtin_ia32_pmulhrsw512_mask ((__v32hi
) __A
,
1226 (__v32hi
) _mm512_setzero_hi(),
1230 static __inline__ __m512i __DEFAULT_FN_ATTRS
1231 _mm512_mulhi_epi16 (__m512i __A
, __m512i __B
)
1233 return (__m512i
) __builtin_ia32_pmulhw512_mask ((__v32hi
) __A
,
1235 (__v32hi
) _mm512_setzero_hi(),
1239 static __inline__ __m512i __DEFAULT_FN_ATTRS
1240 _mm512_mask_mulhi_epi16 (__m512i __W
, __mmask32 __U
, __m512i __A
,
1243 return (__m512i
) __builtin_ia32_pmulhw512_mask ((__v32hi
) __A
,
1249 static __inline__ __m512i __DEFAULT_FN_ATTRS
1250 _mm512_maskz_mulhi_epi16 (__mmask32 __U
, __m512i __A
, __m512i __B
)
1252 return (__m512i
) __builtin_ia32_pmulhw512_mask ((__v32hi
) __A
,
1254 (__v32hi
) _mm512_setzero_hi(),
1258 static __inline__ __m512i __DEFAULT_FN_ATTRS
1259 _mm512_mulhi_epu16 (__m512i __A
, __m512i __B
)
1261 return (__m512i
) __builtin_ia32_pmulhuw512_mask ((__v32hi
) __A
,
1263 (__v32hi
) _mm512_setzero_hi(),
1267 static __inline__ __m512i __DEFAULT_FN_ATTRS
1268 _mm512_mask_mulhi_epu16 (__m512i __W
, __mmask32 __U
, __m512i __A
,
1271 return (__m512i
) __builtin_ia32_pmulhuw512_mask ((__v32hi
) __A
,
1277 static __inline__ __m512i __DEFAULT_FN_ATTRS
1278 _mm512_maskz_mulhi_epu16 (__mmask32 __U
, __m512i __A
, __m512i __B
)
1280 return (__m512i
) __builtin_ia32_pmulhuw512_mask ((__v32hi
) __A
,
1282 (__v32hi
) _mm512_setzero_hi(),
1286 static __inline__ __m512i __DEFAULT_FN_ATTRS
1287 _mm512_maddubs_epi16 (__m512i __X
, __m512i __Y
) {
1288 return (__m512i
) __builtin_ia32_pmaddubsw512_mask ((__v64qi
) __X
,
1290 (__v32hi
) _mm512_setzero_hi(),
1294 static __inline__ __m512i __DEFAULT_FN_ATTRS
1295 _mm512_mask_maddubs_epi16 (__m512i __W
, __mmask32 __U
, __m512i __X
,
1297 return (__m512i
) __builtin_ia32_pmaddubsw512_mask ((__v64qi
) __X
,
1303 static __inline__ __m512i __DEFAULT_FN_ATTRS
1304 _mm512_maskz_maddubs_epi16 (__mmask32 __U
, __m512i __X
, __m512i __Y
) {
1305 return (__m512i
) __builtin_ia32_pmaddubsw512_mask ((__v64qi
) __X
,
1307 (__v32hi
) _mm512_setzero_hi(),
1311 static __inline__ __m512i __DEFAULT_FN_ATTRS
1312 _mm512_madd_epi16 (__m512i __A
, __m512i __B
) {
1313 return (__m512i
) __builtin_ia32_pmaddwd512_mask ((__v32hi
) __A
,
1315 (__v16si
) _mm512_setzero_si512(),
1319 static __inline__ __m512i __DEFAULT_FN_ATTRS
1320 _mm512_mask_madd_epi16 (__m512i __W
, __mmask16 __U
, __m512i __A
,
1322 return (__m512i
) __builtin_ia32_pmaddwd512_mask ((__v32hi
) __A
,
1328 static __inline__ __m512i __DEFAULT_FN_ATTRS
1329 _mm512_maskz_madd_epi16 (__mmask16 __U
, __m512i __A
, __m512i __B
) {
1330 return (__m512i
) __builtin_ia32_pmaddwd512_mask ((__v32hi
) __A
,
1332 (__v16si
) _mm512_setzero_si512(),
1336 static __inline__ __m256i __DEFAULT_FN_ATTRS
1337 _mm512_cvtsepi16_epi8 (__m512i __A
) {
1338 return (__m256i
) __builtin_ia32_pmovswb512_mask ((__v32hi
) __A
,
1339 (__v32qi
)_mm256_setzero_si256(),
1343 static __inline__ __m256i __DEFAULT_FN_ATTRS
1344 _mm512_mask_cvtsepi16_epi8 (__m256i __O
, __mmask32 __M
, __m512i __A
) {
1345 return (__m256i
) __builtin_ia32_pmovswb512_mask ((__v32hi
) __A
,
1350 static __inline__ __m256i __DEFAULT_FN_ATTRS
1351 _mm512_maskz_cvtsepi16_epi8 (__mmask32 __M
, __m512i __A
) {
1352 return (__m256i
) __builtin_ia32_pmovswb512_mask ((__v32hi
) __A
,
1353 (__v32qi
) _mm256_setzero_si256(),
1357 static __inline__ __m256i __DEFAULT_FN_ATTRS
1358 _mm512_cvtusepi16_epi8 (__m512i __A
) {
1359 return (__m256i
) __builtin_ia32_pmovuswb512_mask ((__v32hi
) __A
,
1360 (__v32qi
) _mm256_setzero_si256(),
1364 static __inline__ __m256i __DEFAULT_FN_ATTRS
1365 _mm512_mask_cvtusepi16_epi8 (__m256i __O
, __mmask32 __M
, __m512i __A
) {
1366 return (__m256i
) __builtin_ia32_pmovuswb512_mask ((__v32hi
) __A
,
1371 static __inline__ __m256i __DEFAULT_FN_ATTRS
1372 _mm512_maskz_cvtusepi16_epi8 (__mmask32 __M
, __m512i __A
) {
1373 return (__m256i
) __builtin_ia32_pmovuswb512_mask ((__v32hi
) __A
,
1374 (__v32qi
) _mm256_setzero_si256(),
1378 static __inline__ __m256i __DEFAULT_FN_ATTRS
1379 _mm512_cvtepi16_epi8 (__m512i __A
) {
1380 return (__m256i
) __builtin_ia32_pmovwb512_mask ((__v32hi
) __A
,
1381 (__v32qi
) _mm256_setzero_si256(),
1385 static __inline__ __m256i __DEFAULT_FN_ATTRS
1386 _mm512_mask_cvtepi16_epi8 (__m256i __O
, __mmask32 __M
, __m512i __A
) {
1387 return (__m256i
) __builtin_ia32_pmovwb512_mask ((__v32hi
) __A
,
1392 static __inline__ __m256i __DEFAULT_FN_ATTRS
1393 _mm512_maskz_cvtepi16_epi8 (__mmask32 __M
, __m512i __A
) {
1394 return (__m256i
) __builtin_ia32_pmovwb512_mask ((__v32hi
) __A
,
1395 (__v32qi
) _mm256_setzero_si256(),
1399 static __inline__ __m512i __DEFAULT_FN_ATTRS
1400 _mm512_unpackhi_epi8 (__m512i __A
, __m512i __B
) {
1401 return (__m512i
) __builtin_ia32_punpckhbw512_mask ((__v64qi
) __A
,
1403 (__v64qi
) _mm512_setzero_qi(),
1407 static __inline__ __m512i __DEFAULT_FN_ATTRS
1408 _mm512_mask_unpackhi_epi8 (__m512i __W
, __mmask64 __U
, __m512i __A
,
1410 return (__m512i
) __builtin_ia32_punpckhbw512_mask ((__v64qi
) __A
,
1416 static __inline__ __m512i __DEFAULT_FN_ATTRS
1417 _mm512_maskz_unpackhi_epi8 (__mmask64 __U
, __m512i __A
, __m512i __B
) {
1418 return (__m512i
) __builtin_ia32_punpckhbw512_mask ((__v64qi
) __A
,
1420 (__v64qi
) _mm512_setzero_qi(),
1424 static __inline__ __m512i __DEFAULT_FN_ATTRS
1425 _mm512_unpackhi_epi16 (__m512i __A
, __m512i __B
) {
1426 return (__m512i
) __builtin_ia32_punpckhwd512_mask ((__v32hi
) __A
,
1428 (__v32hi
) _mm512_setzero_hi(),
1432 static __inline__ __m512i __DEFAULT_FN_ATTRS
1433 _mm512_mask_unpackhi_epi16 (__m512i __W
, __mmask32 __U
, __m512i __A
,
1435 return (__m512i
) __builtin_ia32_punpckhwd512_mask ((__v32hi
) __A
,
1441 static __inline__ __m512i __DEFAULT_FN_ATTRS
1442 _mm512_maskz_unpackhi_epi16 (__mmask32 __U
, __m512i __A
, __m512i __B
) {
1443 return (__m512i
) __builtin_ia32_punpckhwd512_mask ((__v32hi
) __A
,
1445 (__v32hi
) _mm512_setzero_hi(),
1449 static __inline__ __m512i __DEFAULT_FN_ATTRS
1450 _mm512_unpacklo_epi8 (__m512i __A
, __m512i __B
) {
1451 return (__m512i
) __builtin_ia32_punpcklbw512_mask ((__v64qi
) __A
,
1453 (__v64qi
) _mm512_setzero_qi(),
1457 static __inline__ __m512i __DEFAULT_FN_ATTRS
1458 _mm512_mask_unpacklo_epi8 (__m512i __W
, __mmask64 __U
, __m512i __A
,
1460 return (__m512i
) __builtin_ia32_punpcklbw512_mask ((__v64qi
) __A
,
1466 static __inline__ __m512i __DEFAULT_FN_ATTRS
1467 _mm512_maskz_unpacklo_epi8 (__mmask64 __U
, __m512i __A
, __m512i __B
) {
1468 return (__m512i
) __builtin_ia32_punpcklbw512_mask ((__v64qi
) __A
,
1470 (__v64qi
) _mm512_setzero_qi(),
1474 static __inline__ __m512i __DEFAULT_FN_ATTRS
1475 _mm512_unpacklo_epi16 (__m512i __A
, __m512i __B
) {
1476 return (__m512i
) __builtin_ia32_punpcklwd512_mask ((__v32hi
) __A
,
1478 (__v32hi
) _mm512_setzero_hi(),
1482 static __inline__ __m512i __DEFAULT_FN_ATTRS
1483 _mm512_mask_unpacklo_epi16 (__m512i __W
, __mmask32 __U
, __m512i __A
,
1485 return (__m512i
) __builtin_ia32_punpcklwd512_mask ((__v32hi
) __A
,
1491 static __inline__ __m512i __DEFAULT_FN_ATTRS
1492 _mm512_maskz_unpacklo_epi16 (__mmask32 __U
, __m512i __A
, __m512i __B
) {
1493 return (__m512i
) __builtin_ia32_punpcklwd512_mask ((__v32hi
) __A
,
1495 (__v32hi
) _mm512_setzero_hi(),
1499 #define _mm512_cmp_epi8_mask(a, b, p) __extension__ ({ \
1500 (__mmask16)__builtin_ia32_cmpb512_mask((__v64qi)(__m512i)(a), \
1501 (__v64qi)(__m512i)(b), \
1502 (p), (__mmask64)-1); })
1504 #define _mm512_mask_cmp_epi8_mask(m, a, b, p) __extension__ ({ \
1505 (__mmask16)__builtin_ia32_cmpb512_mask((__v64qi)(__m512i)(a), \
1506 (__v64qi)(__m512i)(b), \
1507 (p), (__mmask64)(m)); })
1509 #define _mm512_cmp_epu8_mask(a, b, p) __extension__ ({ \
1510 (__mmask16)__builtin_ia32_ucmpb512_mask((__v64qi)(__m512i)(a), \
1511 (__v64qi)(__m512i)(b), \
1512 (p), (__mmask64)-1); })
1514 #define _mm512_mask_cmp_epu8_mask(m, a, b, p) __extension__ ({ \
1515 (__mmask16)__builtin_ia32_ucmpb512_mask((__v64qi)(__m512i)(a), \
1516 (__v64qi)(__m512i)(b), \
1517 (p), (__mmask64)(m)); })
1519 #define _mm512_cmp_epi16_mask(a, b, p) __extension__ ({ \
1520 (__mmask16)__builtin_ia32_cmpw512_mask((__v32hi)(__m512i)(a), \
1521 (__v32hi)(__m512i)(b), \
1522 (p), (__mmask32)-1); })
1524 #define _mm512_mask_cmp_epi16_mask(m, a, b, p) __extension__ ({ \
1525 (__mmask16)__builtin_ia32_cmpw512_mask((__v32hi)(__m512i)(a), \
1526 (__v32hi)(__m512i)(b), \
1527 (p), (__mmask32)(m)); })
1529 #define _mm512_cmp_epu16_mask(a, b, p) __extension__ ({ \
1530 (__mmask16)__builtin_ia32_ucmpw512_mask((__v32hi)(__m512i)(a), \
1531 (__v32hi)(__m512i)(b), \
1532 (p), (__mmask32)-1); })
1534 #define _mm512_mask_cmp_epu16_mask(m, a, b, p) __extension__ ({ \
1535 (__mmask16)__builtin_ia32_ucmpw512_mask((__v32hi)(__m512i)(a), \
1536 (__v32hi)(__m512i)(b), \
1537 (p), (__mmask32)(m)); })
1540 #undef __DEFAULT_FN_ATTRS