1 /*===---- xopintrin.h - XOP intrinsics -------------------------------------===
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:
10 * The above copyright notice and this permission notice shall be included in
11 * all copies or substantial portions of the Software.
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
21 *===-----------------------------------------------------------------------===
25 #error "Never use <xopintrin.h> directly; include <x86intrin.h> instead."
31 #include <fma4intrin.h>
33 /* Define the default attributes for the functions in this file. */
34 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("xop"), __min_vector_width__(128)))
35 #define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("xop"), __min_vector_width__(256)))
37 static __inline__ __m128i __DEFAULT_FN_ATTRS
38 _mm_maccs_epi16(__m128i __A, __m128i __B, __m128i __C)
40 return (__m128i)__builtin_ia32_vpmacssww((__v8hi)__A, (__v8hi)__B, (__v8hi)__C);
43 static __inline__ __m128i __DEFAULT_FN_ATTRS
44 _mm_macc_epi16(__m128i __A, __m128i __B, __m128i __C)
46 return (__m128i)__builtin_ia32_vpmacsww((__v8hi)__A, (__v8hi)__B, (__v8hi)__C);
49 static __inline__ __m128i __DEFAULT_FN_ATTRS
50 _mm_maccsd_epi16(__m128i __A, __m128i __B, __m128i __C)
52 return (__m128i)__builtin_ia32_vpmacsswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
55 static __inline__ __m128i __DEFAULT_FN_ATTRS
56 _mm_maccd_epi16(__m128i __A, __m128i __B, __m128i __C)
58 return (__m128i)__builtin_ia32_vpmacswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
61 static __inline__ __m128i __DEFAULT_FN_ATTRS
62 _mm_maccs_epi32(__m128i __A, __m128i __B, __m128i __C)
64 return (__m128i)__builtin_ia32_vpmacssdd((__v4si)__A, (__v4si)__B, (__v4si)__C);
67 static __inline__ __m128i __DEFAULT_FN_ATTRS
68 _mm_macc_epi32(__m128i __A, __m128i __B, __m128i __C)
70 return (__m128i)__builtin_ia32_vpmacsdd((__v4si)__A, (__v4si)__B, (__v4si)__C);
73 static __inline__ __m128i __DEFAULT_FN_ATTRS
74 _mm_maccslo_epi32(__m128i __A, __m128i __B, __m128i __C)
76 return (__m128i)__builtin_ia32_vpmacssdql((__v4si)__A, (__v4si)__B, (__v2di)__C);
79 static __inline__ __m128i __DEFAULT_FN_ATTRS
80 _mm_macclo_epi32(__m128i __A, __m128i __B, __m128i __C)
82 return (__m128i)__builtin_ia32_vpmacsdql((__v4si)__A, (__v4si)__B, (__v2di)__C);
85 static __inline__ __m128i __DEFAULT_FN_ATTRS
86 _mm_maccshi_epi32(__m128i __A, __m128i __B, __m128i __C)
88 return (__m128i)__builtin_ia32_vpmacssdqh((__v4si)__A, (__v4si)__B, (__v2di)__C);
91 static __inline__ __m128i __DEFAULT_FN_ATTRS
92 _mm_macchi_epi32(__m128i __A, __m128i __B, __m128i __C)
94 return (__m128i)__builtin_ia32_vpmacsdqh((__v4si)__A, (__v4si)__B, (__v2di)__C);
97 static __inline__ __m128i __DEFAULT_FN_ATTRS
98 _mm_maddsd_epi16(__m128i __A, __m128i __B, __m128i __C)
100 return (__m128i)__builtin_ia32_vpmadcsswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
103 static __inline__ __m128i __DEFAULT_FN_ATTRS
104 _mm_maddd_epi16(__m128i __A, __m128i __B, __m128i __C)
106 return (__m128i)__builtin_ia32_vpmadcswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
109 static __inline__ __m128i __DEFAULT_FN_ATTRS
110 _mm_haddw_epi8(__m128i __A)
112 return (__m128i)__builtin_ia32_vphaddbw((__v16qi)__A);
115 static __inline__ __m128i __DEFAULT_FN_ATTRS
116 _mm_haddd_epi8(__m128i __A)
118 return (__m128i)__builtin_ia32_vphaddbd((__v16qi)__A);
121 static __inline__ __m128i __DEFAULT_FN_ATTRS
122 _mm_haddq_epi8(__m128i __A)
124 return (__m128i)__builtin_ia32_vphaddbq((__v16qi)__A);
127 static __inline__ __m128i __DEFAULT_FN_ATTRS
128 _mm_haddd_epi16(__m128i __A)
130 return (__m128i)__builtin_ia32_vphaddwd((__v8hi)__A);
133 static __inline__ __m128i __DEFAULT_FN_ATTRS
134 _mm_haddq_epi16(__m128i __A)
136 return (__m128i)__builtin_ia32_vphaddwq((__v8hi)__A);
139 static __inline__ __m128i __DEFAULT_FN_ATTRS
140 _mm_haddq_epi32(__m128i __A)
142 return (__m128i)__builtin_ia32_vphadddq((__v4si)__A);
145 static __inline__ __m128i __DEFAULT_FN_ATTRS
146 _mm_haddw_epu8(__m128i __A)
148 return (__m128i)__builtin_ia32_vphaddubw((__v16qi)__A);
151 static __inline__ __m128i __DEFAULT_FN_ATTRS
152 _mm_haddd_epu8(__m128i __A)
154 return (__m128i)__builtin_ia32_vphaddubd((__v16qi)__A);
157 static __inline__ __m128i __DEFAULT_FN_ATTRS
158 _mm_haddq_epu8(__m128i __A)
160 return (__m128i)__builtin_ia32_vphaddubq((__v16qi)__A);
163 static __inline__ __m128i __DEFAULT_FN_ATTRS
164 _mm_haddd_epu16(__m128i __A)
166 return (__m128i)__builtin_ia32_vphadduwd((__v8hi)__A);
169 static __inline__ __m128i __DEFAULT_FN_ATTRS
170 _mm_haddq_epu16(__m128i __A)
172 return (__m128i)__builtin_ia32_vphadduwq((__v8hi)__A);
175 static __inline__ __m128i __DEFAULT_FN_ATTRS
176 _mm_haddq_epu32(__m128i __A)
178 return (__m128i)__builtin_ia32_vphaddudq((__v4si)__A);
181 static __inline__ __m128i __DEFAULT_FN_ATTRS
182 _mm_hsubw_epi8(__m128i __A)
184 return (__m128i)__builtin_ia32_vphsubbw((__v16qi)__A);
187 static __inline__ __m128i __DEFAULT_FN_ATTRS
188 _mm_hsubd_epi16(__m128i __A)
190 return (__m128i)__builtin_ia32_vphsubwd((__v8hi)__A);
193 static __inline__ __m128i __DEFAULT_FN_ATTRS
194 _mm_hsubq_epi32(__m128i __A)
196 return (__m128i)__builtin_ia32_vphsubdq((__v4si)__A);
199 static __inline__ __m128i __DEFAULT_FN_ATTRS
200 _mm_cmov_si128(__m128i __A, __m128i __B, __m128i __C)
202 return (__m128i)(((__v2du)__A & (__v2du)__C) | ((__v2du)__B & ~(__v2du)__C));
205 static __inline__ __m256i __DEFAULT_FN_ATTRS256
206 _mm256_cmov_si256(__m256i __A, __m256i __B, __m256i __C)
208 return (__m256i)(((__v4du)__A & (__v4du)__C) | ((__v4du)__B & ~(__v4du)__C));
211 static __inline__ __m128i __DEFAULT_FN_ATTRS
212 _mm_perm_epi8(__m128i __A, __m128i __B, __m128i __C)
214 return (__m128i)__builtin_ia32_vpperm((__v16qi)__A, (__v16qi)__B, (__v16qi)__C);
217 static __inline__ __m128i __DEFAULT_FN_ATTRS
218 _mm_rot_epi8(__m128i __A, __m128i __B)
220 return (__m128i)__builtin_ia32_vprotb((__v16qi)__A, (__v16qi)__B);
223 static __inline__ __m128i __DEFAULT_FN_ATTRS
224 _mm_rot_epi16(__m128i __A, __m128i __B)
226 return (__m128i)__builtin_ia32_vprotw((__v8hi)__A, (__v8hi)__B);
229 static __inline__ __m128i __DEFAULT_FN_ATTRS
230 _mm_rot_epi32(__m128i __A, __m128i __B)
232 return (__m128i)__builtin_ia32_vprotd((__v4si)__A, (__v4si)__B);
235 static __inline__ __m128i __DEFAULT_FN_ATTRS
236 _mm_rot_epi64(__m128i __A, __m128i __B)
238 return (__m128i)__builtin_ia32_vprotq((__v2di)__A, (__v2di)__B);
241 #define _mm_roti_epi8(A, N) \
242 (__m128i)__builtin_ia32_vprotbi((__v16qi)(__m128i)(A), (N))
244 #define _mm_roti_epi16(A, N) \
245 (__m128i)__builtin_ia32_vprotwi((__v8hi)(__m128i)(A), (N))
247 #define _mm_roti_epi32(A, N) \
248 (__m128i)__builtin_ia32_vprotdi((__v4si)(__m128i)(A), (N))
250 #define _mm_roti_epi64(A, N) \
251 (__m128i)__builtin_ia32_vprotqi((__v2di)(__m128i)(A), (N))
253 static __inline__ __m128i __DEFAULT_FN_ATTRS
254 _mm_shl_epi8(__m128i __A, __m128i __B)
256 return (__m128i)__builtin_ia32_vpshlb((__v16qi)__A, (__v16qi)__B);
259 static __inline__ __m128i __DEFAULT_FN_ATTRS
260 _mm_shl_epi16(__m128i __A, __m128i __B)
262 return (__m128i)__builtin_ia32_vpshlw((__v8hi)__A, (__v8hi)__B);
265 static __inline__ __m128i __DEFAULT_FN_ATTRS
266 _mm_shl_epi32(__m128i __A, __m128i __B)
268 return (__m128i)__builtin_ia32_vpshld((__v4si)__A, (__v4si)__B);
271 static __inline__ __m128i __DEFAULT_FN_ATTRS
272 _mm_shl_epi64(__m128i __A, __m128i __B)
274 return (__m128i)__builtin_ia32_vpshlq((__v2di)__A, (__v2di)__B);
277 static __inline__ __m128i __DEFAULT_FN_ATTRS
278 _mm_sha_epi8(__m128i __A, __m128i __B)
280 return (__m128i)__builtin_ia32_vpshab((__v16qi)__A, (__v16qi)__B);
283 static __inline__ __m128i __DEFAULT_FN_ATTRS
284 _mm_sha_epi16(__m128i __A, __m128i __B)
286 return (__m128i)__builtin_ia32_vpshaw((__v8hi)__A, (__v8hi)__B);
289 static __inline__ __m128i __DEFAULT_FN_ATTRS
290 _mm_sha_epi32(__m128i __A, __m128i __B)
292 return (__m128i)__builtin_ia32_vpshad((__v4si)__A, (__v4si)__B);
295 static __inline__ __m128i __DEFAULT_FN_ATTRS
296 _mm_sha_epi64(__m128i __A, __m128i __B)
298 return (__m128i)__builtin_ia32_vpshaq((__v2di)__A, (__v2di)__B);
301 #define _mm_com_epu8(A, B, N) \
302 (__m128i)__builtin_ia32_vpcomub((__v16qi)(__m128i)(A), \
303 (__v16qi)(__m128i)(B), (N))
305 #define _mm_com_epu16(A, B, N) \
306 (__m128i)__builtin_ia32_vpcomuw((__v8hi)(__m128i)(A), \
307 (__v8hi)(__m128i)(B), (N))
309 #define _mm_com_epu32(A, B, N) \
310 (__m128i)__builtin_ia32_vpcomud((__v4si)(__m128i)(A), \
311 (__v4si)(__m128i)(B), (N))
313 #define _mm_com_epu64(A, B, N) \
314 (__m128i)__builtin_ia32_vpcomuq((__v2di)(__m128i)(A), \
315 (__v2di)(__m128i)(B), (N))
317 #define _mm_com_epi8(A, B, N) \
318 (__m128i)__builtin_ia32_vpcomb((__v16qi)(__m128i)(A), \
319 (__v16qi)(__m128i)(B), (N))
321 #define _mm_com_epi16(A, B, N) \
322 (__m128i)__builtin_ia32_vpcomw((__v8hi)(__m128i)(A), \
323 (__v8hi)(__m128i)(B), (N))
325 #define _mm_com_epi32(A, B, N) \
326 (__m128i)__builtin_ia32_vpcomd((__v4si)(__m128i)(A), \
327 (__v4si)(__m128i)(B), (N))
329 #define _mm_com_epi64(A, B, N) \
330 (__m128i)__builtin_ia32_vpcomq((__v2di)(__m128i)(A), \
331 (__v2di)(__m128i)(B), (N))
333 #define _MM_PCOMCTRL_LT 0
334 #define _MM_PCOMCTRL_LE 1
335 #define _MM_PCOMCTRL_GT 2
336 #define _MM_PCOMCTRL_GE 3
337 #define _MM_PCOMCTRL_EQ 4
338 #define _MM_PCOMCTRL_NEQ 5
339 #define _MM_PCOMCTRL_FALSE 6
340 #define _MM_PCOMCTRL_TRUE 7
342 static __inline__ __m128i __DEFAULT_FN_ATTRS
343 _mm_comlt_epu8(__m128i __A, __m128i __B)
345 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_LT);
348 static __inline__ __m128i __DEFAULT_FN_ATTRS
349 _mm_comle_epu8(__m128i __A, __m128i __B)
351 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_LE);
354 static __inline__ __m128i __DEFAULT_FN_ATTRS
355 _mm_comgt_epu8(__m128i __A, __m128i __B)
357 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_GT);
360 static __inline__ __m128i __DEFAULT_FN_ATTRS
361 _mm_comge_epu8(__m128i __A, __m128i __B)
363 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_GE);
366 static __inline__ __m128i __DEFAULT_FN_ATTRS
367 _mm_comeq_epu8(__m128i __A, __m128i __B)
369 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_EQ);
372 static __inline__ __m128i __DEFAULT_FN_ATTRS
373 _mm_comneq_epu8(__m128i __A, __m128i __B)
375 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_NEQ);
378 static __inline__ __m128i __DEFAULT_FN_ATTRS
379 _mm_comfalse_epu8(__m128i __A, __m128i __B)
381 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_FALSE);
384 static __inline__ __m128i __DEFAULT_FN_ATTRS
385 _mm_comtrue_epu8(__m128i __A, __m128i __B)
387 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_TRUE);
390 static __inline__ __m128i __DEFAULT_FN_ATTRS
391 _mm_comlt_epu16(__m128i __A, __m128i __B)
393 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_LT);
396 static __inline__ __m128i __DEFAULT_FN_ATTRS
397 _mm_comle_epu16(__m128i __A, __m128i __B)
399 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_LE);
402 static __inline__ __m128i __DEFAULT_FN_ATTRS
403 _mm_comgt_epu16(__m128i __A, __m128i __B)
405 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_GT);
408 static __inline__ __m128i __DEFAULT_FN_ATTRS
409 _mm_comge_epu16(__m128i __A, __m128i __B)
411 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_GE);
414 static __inline__ __m128i __DEFAULT_FN_ATTRS
415 _mm_comeq_epu16(__m128i __A, __m128i __B)
417 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_EQ);
420 static __inline__ __m128i __DEFAULT_FN_ATTRS
421 _mm_comneq_epu16(__m128i __A, __m128i __B)
423 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_NEQ);
426 static __inline__ __m128i __DEFAULT_FN_ATTRS
427 _mm_comfalse_epu16(__m128i __A, __m128i __B)
429 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_FALSE);
432 static __inline__ __m128i __DEFAULT_FN_ATTRS
433 _mm_comtrue_epu16(__m128i __A, __m128i __B)
435 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_TRUE);
438 static __inline__ __m128i __DEFAULT_FN_ATTRS
439 _mm_comlt_epu32(__m128i __A, __m128i __B)
441 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_LT);
444 static __inline__ __m128i __DEFAULT_FN_ATTRS
445 _mm_comle_epu32(__m128i __A, __m128i __B)
447 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_LE);
450 static __inline__ __m128i __DEFAULT_FN_ATTRS
451 _mm_comgt_epu32(__m128i __A, __m128i __B)
453 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_GT);
456 static __inline__ __m128i __DEFAULT_FN_ATTRS
457 _mm_comge_epu32(__m128i __A, __m128i __B)
459 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_GE);
462 static __inline__ __m128i __DEFAULT_FN_ATTRS
463 _mm_comeq_epu32(__m128i __A, __m128i __B)
465 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_EQ);
468 static __inline__ __m128i __DEFAULT_FN_ATTRS
469 _mm_comneq_epu32(__m128i __A, __m128i __B)
471 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_NEQ);
474 static __inline__ __m128i __DEFAULT_FN_ATTRS
475 _mm_comfalse_epu32(__m128i __A, __m128i __B)
477 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_FALSE);
480 static __inline__ __m128i __DEFAULT_FN_ATTRS
481 _mm_comtrue_epu32(__m128i __A, __m128i __B)
483 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_TRUE);
486 static __inline__ __m128i __DEFAULT_FN_ATTRS
487 _mm_comlt_epu64(__m128i __A, __m128i __B)
489 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_LT);
492 static __inline__ __m128i __DEFAULT_FN_ATTRS
493 _mm_comle_epu64(__m128i __A, __m128i __B)
495 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_LE);
498 static __inline__ __m128i __DEFAULT_FN_ATTRS
499 _mm_comgt_epu64(__m128i __A, __m128i __B)
501 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_GT);
504 static __inline__ __m128i __DEFAULT_FN_ATTRS
505 _mm_comge_epu64(__m128i __A, __m128i __B)
507 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_GE);
510 static __inline__ __m128i __DEFAULT_FN_ATTRS
511 _mm_comeq_epu64(__m128i __A, __m128i __B)
513 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_EQ);
516 static __inline__ __m128i __DEFAULT_FN_ATTRS
517 _mm_comneq_epu64(__m128i __A, __m128i __B)
519 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_NEQ);
522 static __inline__ __m128i __DEFAULT_FN_ATTRS
523 _mm_comfalse_epu64(__m128i __A, __m128i __B)
525 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_FALSE);
528 static __inline__ __m128i __DEFAULT_FN_ATTRS
529 _mm_comtrue_epu64(__m128i __A, __m128i __B)
531 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_TRUE);
534 static __inline__ __m128i __DEFAULT_FN_ATTRS
535 _mm_comlt_epi8(__m128i __A, __m128i __B)
537 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_LT);
540 static __inline__ __m128i __DEFAULT_FN_ATTRS
541 _mm_comle_epi8(__m128i __A, __m128i __B)
543 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_LE);
546 static __inline__ __m128i __DEFAULT_FN_ATTRS
547 _mm_comgt_epi8(__m128i __A, __m128i __B)
549 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_GT);
552 static __inline__ __m128i __DEFAULT_FN_ATTRS
553 _mm_comge_epi8(__m128i __A, __m128i __B)
555 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_GE);
558 static __inline__ __m128i __DEFAULT_FN_ATTRS
559 _mm_comeq_epi8(__m128i __A, __m128i __B)
561 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_EQ);
564 static __inline__ __m128i __DEFAULT_FN_ATTRS
565 _mm_comneq_epi8(__m128i __A, __m128i __B)
567 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_NEQ);
570 static __inline__ __m128i __DEFAULT_FN_ATTRS
571 _mm_comfalse_epi8(__m128i __A, __m128i __B)
573 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_FALSE);
576 static __inline__ __m128i __DEFAULT_FN_ATTRS
577 _mm_comtrue_epi8(__m128i __A, __m128i __B)
579 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_TRUE);
582 static __inline__ __m128i __DEFAULT_FN_ATTRS
583 _mm_comlt_epi16(__m128i __A, __m128i __B)
585 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_LT);
588 static __inline__ __m128i __DEFAULT_FN_ATTRS
589 _mm_comle_epi16(__m128i __A, __m128i __B)
591 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_LE);
594 static __inline__ __m128i __DEFAULT_FN_ATTRS
595 _mm_comgt_epi16(__m128i __A, __m128i __B)
597 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_GT);
600 static __inline__ __m128i __DEFAULT_FN_ATTRS
601 _mm_comge_epi16(__m128i __A, __m128i __B)
603 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_GE);
606 static __inline__ __m128i __DEFAULT_FN_ATTRS
607 _mm_comeq_epi16(__m128i __A, __m128i __B)
609 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_EQ);
612 static __inline__ __m128i __DEFAULT_FN_ATTRS
613 _mm_comneq_epi16(__m128i __A, __m128i __B)
615 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_NEQ);
618 static __inline__ __m128i __DEFAULT_FN_ATTRS
619 _mm_comfalse_epi16(__m128i __A, __m128i __B)
621 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_FALSE);
624 static __inline__ __m128i __DEFAULT_FN_ATTRS
625 _mm_comtrue_epi16(__m128i __A, __m128i __B)
627 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_TRUE);
630 static __inline__ __m128i __DEFAULT_FN_ATTRS
631 _mm_comlt_epi32(__m128i __A, __m128i __B)
633 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_LT);
636 static __inline__ __m128i __DEFAULT_FN_ATTRS
637 _mm_comle_epi32(__m128i __A, __m128i __B)
639 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_LE);
642 static __inline__ __m128i __DEFAULT_FN_ATTRS
643 _mm_comgt_epi32(__m128i __A, __m128i __B)
645 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_GT);
648 static __inline__ __m128i __DEFAULT_FN_ATTRS
649 _mm_comge_epi32(__m128i __A, __m128i __B)
651 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_GE);
654 static __inline__ __m128i __DEFAULT_FN_ATTRS
655 _mm_comeq_epi32(__m128i __A, __m128i __B)
657 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_EQ);
660 static __inline__ __m128i __DEFAULT_FN_ATTRS
661 _mm_comneq_epi32(__m128i __A, __m128i __B)
663 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_NEQ);
666 static __inline__ __m128i __DEFAULT_FN_ATTRS
667 _mm_comfalse_epi32(__m128i __A, __m128i __B)
669 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_FALSE);
672 static __inline__ __m128i __DEFAULT_FN_ATTRS
673 _mm_comtrue_epi32(__m128i __A, __m128i __B)
675 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_TRUE);
678 static __inline__ __m128i __DEFAULT_FN_ATTRS
679 _mm_comlt_epi64(__m128i __A, __m128i __B)
681 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_LT);
684 static __inline__ __m128i __DEFAULT_FN_ATTRS
685 _mm_comle_epi64(__m128i __A, __m128i __B)
687 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_LE);
690 static __inline__ __m128i __DEFAULT_FN_ATTRS
691 _mm_comgt_epi64(__m128i __A, __m128i __B)
693 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_GT);
696 static __inline__ __m128i __DEFAULT_FN_ATTRS
697 _mm_comge_epi64(__m128i __A, __m128i __B)
699 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_GE);
702 static __inline__ __m128i __DEFAULT_FN_ATTRS
703 _mm_comeq_epi64(__m128i __A, __m128i __B)
705 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_EQ);
708 static __inline__ __m128i __DEFAULT_FN_ATTRS
709 _mm_comneq_epi64(__m128i __A, __m128i __B)
711 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_NEQ);
714 static __inline__ __m128i __DEFAULT_FN_ATTRS
715 _mm_comfalse_epi64(__m128i __A, __m128i __B)
717 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_FALSE);
720 static __inline__ __m128i __DEFAULT_FN_ATTRS
721 _mm_comtrue_epi64(__m128i __A, __m128i __B)
723 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_TRUE);
726 #define _mm_permute2_pd(X, Y, C, I) \
727 (__m128d)__builtin_ia32_vpermil2pd((__v2df)(__m128d)(X), \
728 (__v2df)(__m128d)(Y), \
729 (__v2di)(__m128i)(C), (I))
731 #define _mm256_permute2_pd(X, Y, C, I) \
732 (__m256d)__builtin_ia32_vpermil2pd256((__v4df)(__m256d)(X), \
733 (__v4df)(__m256d)(Y), \
734 (__v4di)(__m256i)(C), (I))
736 #define _mm_permute2_ps(X, Y, C, I) \
737 (__m128)__builtin_ia32_vpermil2ps((__v4sf)(__m128)(X), (__v4sf)(__m128)(Y), \
738 (__v4si)(__m128i)(C), (I))
740 #define _mm256_permute2_ps(X, Y, C, I) \
741 (__m256)__builtin_ia32_vpermil2ps256((__v8sf)(__m256)(X), \
742 (__v8sf)(__m256)(Y), \
743 (__v8si)(__m256i)(C), (I))
745 static __inline__ __m128 __DEFAULT_FN_ATTRS
746 _mm_frcz_ss(__m128 __A)
748 return (__m128)__builtin_ia32_vfrczss((__v4sf)__A);
751 static __inline__ __m128d __DEFAULT_FN_ATTRS
752 _mm_frcz_sd(__m128d __A)
754 return (__m128d)__builtin_ia32_vfrczsd((__v2df)__A);
757 static __inline__ __m128 __DEFAULT_FN_ATTRS
758 _mm_frcz_ps(__m128 __A)
760 return (__m128)__builtin_ia32_vfrczps((__v4sf)__A);
763 static __inline__ __m128d __DEFAULT_FN_ATTRS
764 _mm_frcz_pd(__m128d __A)
766 return (__m128d)__builtin_ia32_vfrczpd((__v2df)__A);
769 static __inline__ __m256 __DEFAULT_FN_ATTRS256
770 _mm256_frcz_ps(__m256 __A)
772 return (__m256)__builtin_ia32_vfrczps256((__v8sf)__A);
775 static __inline__ __m256d __DEFAULT_FN_ATTRS256
776 _mm256_frcz_pd(__m256d __A)
778 return (__m256d)__builtin_ia32_vfrczpd256((__v4df)__A);
781 #undef __DEFAULT_FN_ATTRS
782 #undef __DEFAULT_FN_ATTRS256
784 #endif /* __XOPINTRIN_H */