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")))
36 static __inline__ __m128i __DEFAULT_FN_ATTRS
37 _mm_maccs_epi16(__m128i __A, __m128i __B, __m128i __C)
39 return (__m128i)__builtin_ia32_vpmacssww((__v8hi)__A, (__v8hi)__B, (__v8hi)__C);
42 static __inline__ __m128i __DEFAULT_FN_ATTRS
43 _mm_macc_epi16(__m128i __A, __m128i __B, __m128i __C)
45 return (__m128i)__builtin_ia32_vpmacsww((__v8hi)__A, (__v8hi)__B, (__v8hi)__C);
48 static __inline__ __m128i __DEFAULT_FN_ATTRS
49 _mm_maccsd_epi16(__m128i __A, __m128i __B, __m128i __C)
51 return (__m128i)__builtin_ia32_vpmacsswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
54 static __inline__ __m128i __DEFAULT_FN_ATTRS
55 _mm_maccd_epi16(__m128i __A, __m128i __B, __m128i __C)
57 return (__m128i)__builtin_ia32_vpmacswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
60 static __inline__ __m128i __DEFAULT_FN_ATTRS
61 _mm_maccs_epi32(__m128i __A, __m128i __B, __m128i __C)
63 return (__m128i)__builtin_ia32_vpmacssdd((__v4si)__A, (__v4si)__B, (__v4si)__C);
66 static __inline__ __m128i __DEFAULT_FN_ATTRS
67 _mm_macc_epi32(__m128i __A, __m128i __B, __m128i __C)
69 return (__m128i)__builtin_ia32_vpmacsdd((__v4si)__A, (__v4si)__B, (__v4si)__C);
72 static __inline__ __m128i __DEFAULT_FN_ATTRS
73 _mm_maccslo_epi32(__m128i __A, __m128i __B, __m128i __C)
75 return (__m128i)__builtin_ia32_vpmacssdql((__v4si)__A, (__v4si)__B, (__v2di)__C);
78 static __inline__ __m128i __DEFAULT_FN_ATTRS
79 _mm_macclo_epi32(__m128i __A, __m128i __B, __m128i __C)
81 return (__m128i)__builtin_ia32_vpmacsdql((__v4si)__A, (__v4si)__B, (__v2di)__C);
84 static __inline__ __m128i __DEFAULT_FN_ATTRS
85 _mm_maccshi_epi32(__m128i __A, __m128i __B, __m128i __C)
87 return (__m128i)__builtin_ia32_vpmacssdqh((__v4si)__A, (__v4si)__B, (__v2di)__C);
90 static __inline__ __m128i __DEFAULT_FN_ATTRS
91 _mm_macchi_epi32(__m128i __A, __m128i __B, __m128i __C)
93 return (__m128i)__builtin_ia32_vpmacsdqh((__v4si)__A, (__v4si)__B, (__v2di)__C);
96 static __inline__ __m128i __DEFAULT_FN_ATTRS
97 _mm_maddsd_epi16(__m128i __A, __m128i __B, __m128i __C)
99 return (__m128i)__builtin_ia32_vpmadcsswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
102 static __inline__ __m128i __DEFAULT_FN_ATTRS
103 _mm_maddd_epi16(__m128i __A, __m128i __B, __m128i __C)
105 return (__m128i)__builtin_ia32_vpmadcswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
108 static __inline__ __m128i __DEFAULT_FN_ATTRS
109 _mm_haddw_epi8(__m128i __A)
111 return (__m128i)__builtin_ia32_vphaddbw((__v16qi)__A);
114 static __inline__ __m128i __DEFAULT_FN_ATTRS
115 _mm_haddd_epi8(__m128i __A)
117 return (__m128i)__builtin_ia32_vphaddbd((__v16qi)__A);
120 static __inline__ __m128i __DEFAULT_FN_ATTRS
121 _mm_haddq_epi8(__m128i __A)
123 return (__m128i)__builtin_ia32_vphaddbq((__v16qi)__A);
126 static __inline__ __m128i __DEFAULT_FN_ATTRS
127 _mm_haddd_epi16(__m128i __A)
129 return (__m128i)__builtin_ia32_vphaddwd((__v8hi)__A);
132 static __inline__ __m128i __DEFAULT_FN_ATTRS
133 _mm_haddq_epi16(__m128i __A)
135 return (__m128i)__builtin_ia32_vphaddwq((__v8hi)__A);
138 static __inline__ __m128i __DEFAULT_FN_ATTRS
139 _mm_haddq_epi32(__m128i __A)
141 return (__m128i)__builtin_ia32_vphadddq((__v4si)__A);
144 static __inline__ __m128i __DEFAULT_FN_ATTRS
145 _mm_haddw_epu8(__m128i __A)
147 return (__m128i)__builtin_ia32_vphaddubw((__v16qi)__A);
150 static __inline__ __m128i __DEFAULT_FN_ATTRS
151 _mm_haddd_epu8(__m128i __A)
153 return (__m128i)__builtin_ia32_vphaddubd((__v16qi)__A);
156 static __inline__ __m128i __DEFAULT_FN_ATTRS
157 _mm_haddq_epu8(__m128i __A)
159 return (__m128i)__builtin_ia32_vphaddubq((__v16qi)__A);
162 static __inline__ __m128i __DEFAULT_FN_ATTRS
163 _mm_haddd_epu16(__m128i __A)
165 return (__m128i)__builtin_ia32_vphadduwd((__v8hi)__A);
168 static __inline__ __m128i __DEFAULT_FN_ATTRS
169 _mm_haddq_epu16(__m128i __A)
171 return (__m128i)__builtin_ia32_vphadduwq((__v8hi)__A);
174 static __inline__ __m128i __DEFAULT_FN_ATTRS
175 _mm_haddq_epu32(__m128i __A)
177 return (__m128i)__builtin_ia32_vphaddudq((__v4si)__A);
180 static __inline__ __m128i __DEFAULT_FN_ATTRS
181 _mm_hsubw_epi8(__m128i __A)
183 return (__m128i)__builtin_ia32_vphsubbw((__v16qi)__A);
186 static __inline__ __m128i __DEFAULT_FN_ATTRS
187 _mm_hsubd_epi16(__m128i __A)
189 return (__m128i)__builtin_ia32_vphsubwd((__v8hi)__A);
192 static __inline__ __m128i __DEFAULT_FN_ATTRS
193 _mm_hsubq_epi32(__m128i __A)
195 return (__m128i)__builtin_ia32_vphsubdq((__v4si)__A);
198 static __inline__ __m128i __DEFAULT_FN_ATTRS
199 _mm_cmov_si128(__m128i __A, __m128i __B, __m128i __C)
201 return (__m128i)__builtin_ia32_vpcmov(__A, __B, __C);
204 static __inline__ __m256i __DEFAULT_FN_ATTRS
205 _mm256_cmov_si256(__m256i __A, __m256i __B, __m256i __C)
207 return (__m256i)__builtin_ia32_vpcmov_256(__A, __B, __C);
210 static __inline__ __m128i __DEFAULT_FN_ATTRS
211 _mm_perm_epi8(__m128i __A, __m128i __B, __m128i __C)
213 return (__m128i)__builtin_ia32_vpperm((__v16qi)__A, (__v16qi)__B, (__v16qi)__C);
216 static __inline__ __m128i __DEFAULT_FN_ATTRS
217 _mm_rot_epi8(__m128i __A, __m128i __B)
219 return (__m128i)__builtin_ia32_vprotb((__v16qi)__A, (__v16qi)__B);
222 static __inline__ __m128i __DEFAULT_FN_ATTRS
223 _mm_rot_epi16(__m128i __A, __m128i __B)
225 return (__m128i)__builtin_ia32_vprotw((__v8hi)__A, (__v8hi)__B);
228 static __inline__ __m128i __DEFAULT_FN_ATTRS
229 _mm_rot_epi32(__m128i __A, __m128i __B)
231 return (__m128i)__builtin_ia32_vprotd((__v4si)__A, (__v4si)__B);
234 static __inline__ __m128i __DEFAULT_FN_ATTRS
235 _mm_rot_epi64(__m128i __A, __m128i __B)
237 return (__m128i)__builtin_ia32_vprotq((__v2di)__A, (__v2di)__B);
240 #define _mm_roti_epi8(A, N) __extension__ ({ \
242 (__m128i)__builtin_ia32_vprotbi((__v16qi)__A, (N)); })
244 #define _mm_roti_epi16(A, N) __extension__ ({ \
246 (__m128i)__builtin_ia32_vprotwi((__v8hi)__A, (N)); })
248 #define _mm_roti_epi32(A, N) __extension__ ({ \
250 (__m128i)__builtin_ia32_vprotdi((__v4si)__A, (N)); })
252 #define _mm_roti_epi64(A, N) __extension__ ({ \
254 (__m128i)__builtin_ia32_vprotqi((__v2di)__A, (N)); })
256 static __inline__ __m128i __DEFAULT_FN_ATTRS
257 _mm_shl_epi8(__m128i __A, __m128i __B)
259 return (__m128i)__builtin_ia32_vpshlb((__v16qi)__A, (__v16qi)__B);
262 static __inline__ __m128i __DEFAULT_FN_ATTRS
263 _mm_shl_epi16(__m128i __A, __m128i __B)
265 return (__m128i)__builtin_ia32_vpshlw((__v8hi)__A, (__v8hi)__B);
268 static __inline__ __m128i __DEFAULT_FN_ATTRS
269 _mm_shl_epi32(__m128i __A, __m128i __B)
271 return (__m128i)__builtin_ia32_vpshld((__v4si)__A, (__v4si)__B);
274 static __inline__ __m128i __DEFAULT_FN_ATTRS
275 _mm_shl_epi64(__m128i __A, __m128i __B)
277 return (__m128i)__builtin_ia32_vpshlq((__v2di)__A, (__v2di)__B);
280 static __inline__ __m128i __DEFAULT_FN_ATTRS
281 _mm_sha_epi8(__m128i __A, __m128i __B)
283 return (__m128i)__builtin_ia32_vpshab((__v16qi)__A, (__v16qi)__B);
286 static __inline__ __m128i __DEFAULT_FN_ATTRS
287 _mm_sha_epi16(__m128i __A, __m128i __B)
289 return (__m128i)__builtin_ia32_vpshaw((__v8hi)__A, (__v8hi)__B);
292 static __inline__ __m128i __DEFAULT_FN_ATTRS
293 _mm_sha_epi32(__m128i __A, __m128i __B)
295 return (__m128i)__builtin_ia32_vpshad((__v4si)__A, (__v4si)__B);
298 static __inline__ __m128i __DEFAULT_FN_ATTRS
299 _mm_sha_epi64(__m128i __A, __m128i __B)
301 return (__m128i)__builtin_ia32_vpshaq((__v2di)__A, (__v2di)__B);
304 #define _mm_com_epu8(A, B, N) __extension__ ({ \
307 (__m128i)__builtin_ia32_vpcomub((__v16qi)__A, (__v16qi)__B, (N)); })
309 #define _mm_com_epu16(A, B, N) __extension__ ({ \
312 (__m128i)__builtin_ia32_vpcomuw((__v8hi)__A, (__v8hi)__B, (N)); })
314 #define _mm_com_epu32(A, B, N) __extension__ ({ \
317 (__m128i)__builtin_ia32_vpcomud((__v4si)__A, (__v4si)__B, (N)); })
319 #define _mm_com_epu64(A, B, N) __extension__ ({ \
322 (__m128i)__builtin_ia32_vpcomuq((__v2di)__A, (__v2di)__B, (N)); })
324 #define _mm_com_epi8(A, B, N) __extension__ ({ \
327 (__m128i)__builtin_ia32_vpcomb((__v16qi)__A, (__v16qi)__B, (N)); })
329 #define _mm_com_epi16(A, B, N) __extension__ ({ \
332 (__m128i)__builtin_ia32_vpcomw((__v8hi)__A, (__v8hi)__B, (N)); })
334 #define _mm_com_epi32(A, B, N) __extension__ ({ \
337 (__m128i)__builtin_ia32_vpcomd((__v4si)__A, (__v4si)__B, (N)); })
339 #define _mm_com_epi64(A, B, N) __extension__ ({ \
342 (__m128i)__builtin_ia32_vpcomq((__v2di)__A, (__v2di)__B, (N)); })
344 #define _MM_PCOMCTRL_LT 0
345 #define _MM_PCOMCTRL_LE 1
346 #define _MM_PCOMCTRL_GT 2
347 #define _MM_PCOMCTRL_GE 3
348 #define _MM_PCOMCTRL_EQ 4
349 #define _MM_PCOMCTRL_NEQ 5
350 #define _MM_PCOMCTRL_FALSE 6
351 #define _MM_PCOMCTRL_TRUE 7
353 static __inline__ __m128i __DEFAULT_FN_ATTRS
354 _mm_comlt_epu8(__m128i __A, __m128i __B)
356 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_LT);
359 static __inline__ __m128i __DEFAULT_FN_ATTRS
360 _mm_comle_epu8(__m128i __A, __m128i __B)
362 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_LE);
365 static __inline__ __m128i __DEFAULT_FN_ATTRS
366 _mm_comgt_epu8(__m128i __A, __m128i __B)
368 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_GT);
371 static __inline__ __m128i __DEFAULT_FN_ATTRS
372 _mm_comge_epu8(__m128i __A, __m128i __B)
374 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_GE);
377 static __inline__ __m128i __DEFAULT_FN_ATTRS
378 _mm_comeq_epu8(__m128i __A, __m128i __B)
380 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_EQ);
383 static __inline__ __m128i __DEFAULT_FN_ATTRS
384 _mm_comneq_epu8(__m128i __A, __m128i __B)
386 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_NEQ);
389 static __inline__ __m128i __DEFAULT_FN_ATTRS
390 _mm_comfalse_epu8(__m128i __A, __m128i __B)
392 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_FALSE);
395 static __inline__ __m128i __DEFAULT_FN_ATTRS
396 _mm_comtrue_epu8(__m128i __A, __m128i __B)
398 return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_TRUE);
401 static __inline__ __m128i __DEFAULT_FN_ATTRS
402 _mm_comlt_epu16(__m128i __A, __m128i __B)
404 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_LT);
407 static __inline__ __m128i __DEFAULT_FN_ATTRS
408 _mm_comle_epu16(__m128i __A, __m128i __B)
410 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_LE);
413 static __inline__ __m128i __DEFAULT_FN_ATTRS
414 _mm_comgt_epu16(__m128i __A, __m128i __B)
416 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_GT);
419 static __inline__ __m128i __DEFAULT_FN_ATTRS
420 _mm_comge_epu16(__m128i __A, __m128i __B)
422 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_GE);
425 static __inline__ __m128i __DEFAULT_FN_ATTRS
426 _mm_comeq_epu16(__m128i __A, __m128i __B)
428 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_EQ);
431 static __inline__ __m128i __DEFAULT_FN_ATTRS
432 _mm_comneq_epu16(__m128i __A, __m128i __B)
434 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_NEQ);
437 static __inline__ __m128i __DEFAULT_FN_ATTRS
438 _mm_comfalse_epu16(__m128i __A, __m128i __B)
440 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_FALSE);
443 static __inline__ __m128i __DEFAULT_FN_ATTRS
444 _mm_comtrue_epu16(__m128i __A, __m128i __B)
446 return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_TRUE);
449 static __inline__ __m128i __DEFAULT_FN_ATTRS
450 _mm_comlt_epu32(__m128i __A, __m128i __B)
452 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_LT);
455 static __inline__ __m128i __DEFAULT_FN_ATTRS
456 _mm_comle_epu32(__m128i __A, __m128i __B)
458 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_LE);
461 static __inline__ __m128i __DEFAULT_FN_ATTRS
462 _mm_comgt_epu32(__m128i __A, __m128i __B)
464 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_GT);
467 static __inline__ __m128i __DEFAULT_FN_ATTRS
468 _mm_comge_epu32(__m128i __A, __m128i __B)
470 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_GE);
473 static __inline__ __m128i __DEFAULT_FN_ATTRS
474 _mm_comeq_epu32(__m128i __A, __m128i __B)
476 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_EQ);
479 static __inline__ __m128i __DEFAULT_FN_ATTRS
480 _mm_comneq_epu32(__m128i __A, __m128i __B)
482 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_NEQ);
485 static __inline__ __m128i __DEFAULT_FN_ATTRS
486 _mm_comfalse_epu32(__m128i __A, __m128i __B)
488 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_FALSE);
491 static __inline__ __m128i __DEFAULT_FN_ATTRS
492 _mm_comtrue_epu32(__m128i __A, __m128i __B)
494 return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_TRUE);
497 static __inline__ __m128i __DEFAULT_FN_ATTRS
498 _mm_comlt_epu64(__m128i __A, __m128i __B)
500 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_LT);
503 static __inline__ __m128i __DEFAULT_FN_ATTRS
504 _mm_comle_epu64(__m128i __A, __m128i __B)
506 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_LE);
509 static __inline__ __m128i __DEFAULT_FN_ATTRS
510 _mm_comgt_epu64(__m128i __A, __m128i __B)
512 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_GT);
515 static __inline__ __m128i __DEFAULT_FN_ATTRS
516 _mm_comge_epu64(__m128i __A, __m128i __B)
518 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_GE);
521 static __inline__ __m128i __DEFAULT_FN_ATTRS
522 _mm_comeq_epu64(__m128i __A, __m128i __B)
524 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_EQ);
527 static __inline__ __m128i __DEFAULT_FN_ATTRS
528 _mm_comneq_epu64(__m128i __A, __m128i __B)
530 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_NEQ);
533 static __inline__ __m128i __DEFAULT_FN_ATTRS
534 _mm_comfalse_epu64(__m128i __A, __m128i __B)
536 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_FALSE);
539 static __inline__ __m128i __DEFAULT_FN_ATTRS
540 _mm_comtrue_epu64(__m128i __A, __m128i __B)
542 return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_TRUE);
545 static __inline__ __m128i __DEFAULT_FN_ATTRS
546 _mm_comlt_epi8(__m128i __A, __m128i __B)
548 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_LT);
551 static __inline__ __m128i __DEFAULT_FN_ATTRS
552 _mm_comle_epi8(__m128i __A, __m128i __B)
554 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_LE);
557 static __inline__ __m128i __DEFAULT_FN_ATTRS
558 _mm_comgt_epi8(__m128i __A, __m128i __B)
560 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_GT);
563 static __inline__ __m128i __DEFAULT_FN_ATTRS
564 _mm_comge_epi8(__m128i __A, __m128i __B)
566 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_GE);
569 static __inline__ __m128i __DEFAULT_FN_ATTRS
570 _mm_comeq_epi8(__m128i __A, __m128i __B)
572 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_EQ);
575 static __inline__ __m128i __DEFAULT_FN_ATTRS
576 _mm_comneq_epi8(__m128i __A, __m128i __B)
578 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_NEQ);
581 static __inline__ __m128i __DEFAULT_FN_ATTRS
582 _mm_comfalse_epi8(__m128i __A, __m128i __B)
584 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_FALSE);
587 static __inline__ __m128i __DEFAULT_FN_ATTRS
588 _mm_comtrue_epi8(__m128i __A, __m128i __B)
590 return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_TRUE);
593 static __inline__ __m128i __DEFAULT_FN_ATTRS
594 _mm_comlt_epi16(__m128i __A, __m128i __B)
596 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_LT);
599 static __inline__ __m128i __DEFAULT_FN_ATTRS
600 _mm_comle_epi16(__m128i __A, __m128i __B)
602 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_LE);
605 static __inline__ __m128i __DEFAULT_FN_ATTRS
606 _mm_comgt_epi16(__m128i __A, __m128i __B)
608 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_GT);
611 static __inline__ __m128i __DEFAULT_FN_ATTRS
612 _mm_comge_epi16(__m128i __A, __m128i __B)
614 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_GE);
617 static __inline__ __m128i __DEFAULT_FN_ATTRS
618 _mm_comeq_epi16(__m128i __A, __m128i __B)
620 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_EQ);
623 static __inline__ __m128i __DEFAULT_FN_ATTRS
624 _mm_comneq_epi16(__m128i __A, __m128i __B)
626 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_NEQ);
629 static __inline__ __m128i __DEFAULT_FN_ATTRS
630 _mm_comfalse_epi16(__m128i __A, __m128i __B)
632 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_FALSE);
635 static __inline__ __m128i __DEFAULT_FN_ATTRS
636 _mm_comtrue_epi16(__m128i __A, __m128i __B)
638 return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_TRUE);
641 static __inline__ __m128i __DEFAULT_FN_ATTRS
642 _mm_comlt_epi32(__m128i __A, __m128i __B)
644 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_LT);
647 static __inline__ __m128i __DEFAULT_FN_ATTRS
648 _mm_comle_epi32(__m128i __A, __m128i __B)
650 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_LE);
653 static __inline__ __m128i __DEFAULT_FN_ATTRS
654 _mm_comgt_epi32(__m128i __A, __m128i __B)
656 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_GT);
659 static __inline__ __m128i __DEFAULT_FN_ATTRS
660 _mm_comge_epi32(__m128i __A, __m128i __B)
662 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_GE);
665 static __inline__ __m128i __DEFAULT_FN_ATTRS
666 _mm_comeq_epi32(__m128i __A, __m128i __B)
668 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_EQ);
671 static __inline__ __m128i __DEFAULT_FN_ATTRS
672 _mm_comneq_epi32(__m128i __A, __m128i __B)
674 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_NEQ);
677 static __inline__ __m128i __DEFAULT_FN_ATTRS
678 _mm_comfalse_epi32(__m128i __A, __m128i __B)
680 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_FALSE);
683 static __inline__ __m128i __DEFAULT_FN_ATTRS
684 _mm_comtrue_epi32(__m128i __A, __m128i __B)
686 return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_TRUE);
689 static __inline__ __m128i __DEFAULT_FN_ATTRS
690 _mm_comlt_epi64(__m128i __A, __m128i __B)
692 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_LT);
695 static __inline__ __m128i __DEFAULT_FN_ATTRS
696 _mm_comle_epi64(__m128i __A, __m128i __B)
698 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_LE);
701 static __inline__ __m128i __DEFAULT_FN_ATTRS
702 _mm_comgt_epi64(__m128i __A, __m128i __B)
704 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_GT);
707 static __inline__ __m128i __DEFAULT_FN_ATTRS
708 _mm_comge_epi64(__m128i __A, __m128i __B)
710 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_GE);
713 static __inline__ __m128i __DEFAULT_FN_ATTRS
714 _mm_comeq_epi64(__m128i __A, __m128i __B)
716 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_EQ);
719 static __inline__ __m128i __DEFAULT_FN_ATTRS
720 _mm_comneq_epi64(__m128i __A, __m128i __B)
722 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_NEQ);
725 static __inline__ __m128i __DEFAULT_FN_ATTRS
726 _mm_comfalse_epi64(__m128i __A, __m128i __B)
728 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_FALSE);
731 static __inline__ __m128i __DEFAULT_FN_ATTRS
732 _mm_comtrue_epi64(__m128i __A, __m128i __B)
734 return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_TRUE);
737 #define _mm_permute2_pd(X, Y, C, I) __extension__ ({ \
741 (__m128d)__builtin_ia32_vpermil2pd((__v2df)__X, (__v2df)__Y, \
742 (__v2di)__C, (I)); })
744 #define _mm256_permute2_pd(X, Y, C, I) __extension__ ({ \
748 (__m256d)__builtin_ia32_vpermil2pd256((__v4df)__X, (__v4df)__Y, \
749 (__v4di)__C, (I)); })
751 #define _mm_permute2_ps(X, Y, C, I) __extension__ ({ \
755 (__m128)__builtin_ia32_vpermil2ps((__v4sf)__X, (__v4sf)__Y, \
756 (__v4si)__C, (I)); })
758 #define _mm256_permute2_ps(X, Y, C, I) __extension__ ({ \
762 (__m256)__builtin_ia32_vpermil2ps256((__v8sf)__X, (__v8sf)__Y, \
763 (__v8si)__C, (I)); })
765 static __inline__ __m128 __DEFAULT_FN_ATTRS
766 _mm_frcz_ss(__m128 __A)
768 return (__m128)__builtin_ia32_vfrczss((__v4sf)__A);
771 static __inline__ __m128d __DEFAULT_FN_ATTRS
772 _mm_frcz_sd(__m128d __A)
774 return (__m128d)__builtin_ia32_vfrczsd((__v2df)__A);
777 static __inline__ __m128 __DEFAULT_FN_ATTRS
778 _mm_frcz_ps(__m128 __A)
780 return (__m128)__builtin_ia32_vfrczps((__v4sf)__A);
783 static __inline__ __m128d __DEFAULT_FN_ATTRS
784 _mm_frcz_pd(__m128d __A)
786 return (__m128d)__builtin_ia32_vfrczpd((__v2df)__A);
789 static __inline__ __m256 __DEFAULT_FN_ATTRS
790 _mm256_frcz_ps(__m256 __A)
792 return (__m256)__builtin_ia32_vfrczps256((__v8sf)__A);
795 static __inline__ __m256d __DEFAULT_FN_ATTRS
796 _mm256_frcz_pd(__m256d __A)
798 return (__m256d)__builtin_ia32_vfrczpd256((__v4df)__A);
801 #undef __DEFAULT_FN_ATTRS
803 #endif /* __XOPINTRIN_H */