1/*===---- xopintrin.h - XOP 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 __X86INTRIN_H
25#error "Never use <xopintrin.h> directly; include <x86intrin.h> instead."
26#endif
27
28#ifndef __XOPINTRIN_H
29#define __XOPINTRIN_H
30
31#include <fma4intrin.h>
32
33/* Define the default attributes for the functions in this file. */
34#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("xop")))
35
36static __inline__ __m128i __DEFAULT_FN_ATTRS
37_mm_maccs_epi16(__m128i __A, __m128i __B, __m128i __C)
38{
39  return (__m128i)__builtin_ia32_vpmacssww((__v8hi)__A, (__v8hi)__B, (__v8hi)__C);
40}
41
42static __inline__ __m128i __DEFAULT_FN_ATTRS
43_mm_macc_epi16(__m128i __A, __m128i __B, __m128i __C)
44{
45  return (__m128i)__builtin_ia32_vpmacsww((__v8hi)__A, (__v8hi)__B, (__v8hi)__C);
46}
47
48static __inline__ __m128i __DEFAULT_FN_ATTRS
49_mm_maccsd_epi16(__m128i __A, __m128i __B, __m128i __C)
50{
51  return (__m128i)__builtin_ia32_vpmacsswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
52}
53
54static __inline__ __m128i __DEFAULT_FN_ATTRS
55_mm_maccd_epi16(__m128i __A, __m128i __B, __m128i __C)
56{
57  return (__m128i)__builtin_ia32_vpmacswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
58}
59
60static __inline__ __m128i __DEFAULT_FN_ATTRS
61_mm_maccs_epi32(__m128i __A, __m128i __B, __m128i __C)
62{
63  return (__m128i)__builtin_ia32_vpmacssdd((__v4si)__A, (__v4si)__B, (__v4si)__C);
64}
65
66static __inline__ __m128i __DEFAULT_FN_ATTRS
67_mm_macc_epi32(__m128i __A, __m128i __B, __m128i __C)
68{
69  return (__m128i)__builtin_ia32_vpmacsdd((__v4si)__A, (__v4si)__B, (__v4si)__C);
70}
71
72static __inline__ __m128i __DEFAULT_FN_ATTRS
73_mm_maccslo_epi32(__m128i __A, __m128i __B, __m128i __C)
74{
75  return (__m128i)__builtin_ia32_vpmacssdql((__v4si)__A, (__v4si)__B, (__v2di)__C);
76}
77
78static __inline__ __m128i __DEFAULT_FN_ATTRS
79_mm_macclo_epi32(__m128i __A, __m128i __B, __m128i __C)
80{
81  return (__m128i)__builtin_ia32_vpmacsdql((__v4si)__A, (__v4si)__B, (__v2di)__C);
82}
83
84static __inline__ __m128i __DEFAULT_FN_ATTRS
85_mm_maccshi_epi32(__m128i __A, __m128i __B, __m128i __C)
86{
87  return (__m128i)__builtin_ia32_vpmacssdqh((__v4si)__A, (__v4si)__B, (__v2di)__C);
88}
89
90static __inline__ __m128i __DEFAULT_FN_ATTRS
91_mm_macchi_epi32(__m128i __A, __m128i __B, __m128i __C)
92{
93  return (__m128i)__builtin_ia32_vpmacsdqh((__v4si)__A, (__v4si)__B, (__v2di)__C);
94}
95
96static __inline__ __m128i __DEFAULT_FN_ATTRS
97_mm_maddsd_epi16(__m128i __A, __m128i __B, __m128i __C)
98{
99  return (__m128i)__builtin_ia32_vpmadcsswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
100}
101
102static __inline__ __m128i __DEFAULT_FN_ATTRS
103_mm_maddd_epi16(__m128i __A, __m128i __B, __m128i __C)
104{
105  return (__m128i)__builtin_ia32_vpmadcswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
106}
107
108static __inline__ __m128i __DEFAULT_FN_ATTRS
109_mm_haddw_epi8(__m128i __A)
110{
111  return (__m128i)__builtin_ia32_vphaddbw((__v16qi)__A);
112}
113
114static __inline__ __m128i __DEFAULT_FN_ATTRS
115_mm_haddd_epi8(__m128i __A)
116{
117  return (__m128i)__builtin_ia32_vphaddbd((__v16qi)__A);
118}
119
120static __inline__ __m128i __DEFAULT_FN_ATTRS
121_mm_haddq_epi8(__m128i __A)
122{
123  return (__m128i)__builtin_ia32_vphaddbq((__v16qi)__A);
124}
125
126static __inline__ __m128i __DEFAULT_FN_ATTRS
127_mm_haddd_epi16(__m128i __A)
128{
129  return (__m128i)__builtin_ia32_vphaddwd((__v8hi)__A);
130}
131
132static __inline__ __m128i __DEFAULT_FN_ATTRS
133_mm_haddq_epi16(__m128i __A)
134{
135  return (__m128i)__builtin_ia32_vphaddwq((__v8hi)__A);
136}
137
138static __inline__ __m128i __DEFAULT_FN_ATTRS
139_mm_haddq_epi32(__m128i __A)
140{
141  return (__m128i)__builtin_ia32_vphadddq((__v4si)__A);
142}
143
144static __inline__ __m128i __DEFAULT_FN_ATTRS
145_mm_haddw_epu8(__m128i __A)
146{
147  return (__m128i)__builtin_ia32_vphaddubw((__v16qi)__A);
148}
149
150static __inline__ __m128i __DEFAULT_FN_ATTRS
151_mm_haddd_epu8(__m128i __A)
152{
153  return (__m128i)__builtin_ia32_vphaddubd((__v16qi)__A);
154}
155
156static __inline__ __m128i __DEFAULT_FN_ATTRS
157_mm_haddq_epu8(__m128i __A)
158{
159  return (__m128i)__builtin_ia32_vphaddubq((__v16qi)__A);
160}
161
162static __inline__ __m128i __DEFAULT_FN_ATTRS
163_mm_haddd_epu16(__m128i __A)
164{
165  return (__m128i)__builtin_ia32_vphadduwd((__v8hi)__A);
166}
167
168static __inline__ __m128i __DEFAULT_FN_ATTRS
169_mm_haddq_epu16(__m128i __A)
170{
171  return (__m128i)__builtin_ia32_vphadduwq((__v8hi)__A);
172}
173
174static __inline__ __m128i __DEFAULT_FN_ATTRS
175_mm_haddq_epu32(__m128i __A)
176{
177  return (__m128i)__builtin_ia32_vphaddudq((__v4si)__A);
178}
179
180static __inline__ __m128i __DEFAULT_FN_ATTRS
181_mm_hsubw_epi8(__m128i __A)
182{
183  return (__m128i)__builtin_ia32_vphsubbw((__v16qi)__A);
184}
185
186static __inline__ __m128i __DEFAULT_FN_ATTRS
187_mm_hsubd_epi16(__m128i __A)
188{
189  return (__m128i)__builtin_ia32_vphsubwd((__v8hi)__A);
190}
191
192static __inline__ __m128i __DEFAULT_FN_ATTRS
193_mm_hsubq_epi32(__m128i __A)
194{
195  return (__m128i)__builtin_ia32_vphsubdq((__v4si)__A);
196}
197
198static __inline__ __m128i __DEFAULT_FN_ATTRS
199_mm_cmov_si128(__m128i __A, __m128i __B, __m128i __C)
200{
201  return (__m128i)(((__v2du)__A & (__v2du)__C) | ((__v2du)__B & ~(__v2du)__C));
202}
203
204static __inline__ __m256i __DEFAULT_FN_ATTRS
205_mm256_cmov_si256(__m256i __A, __m256i __B, __m256i __C)
206{
207  return (__m256i)(((__v4du)__A & (__v4du)__C) | ((__v4du)__B & ~(__v4du)__C));
208}
209
210static __inline__ __m128i __DEFAULT_FN_ATTRS
211_mm_perm_epi8(__m128i __A, __m128i __B, __m128i __C)
212{
213  return (__m128i)__builtin_ia32_vpperm((__v16qi)__A, (__v16qi)__B, (__v16qi)__C);
214}
215
216static __inline__ __m128i __DEFAULT_FN_ATTRS
217_mm_rot_epi8(__m128i __A, __m128i __B)
218{
219  return (__m128i)__builtin_ia32_vprotb((__v16qi)__A, (__v16qi)__B);
220}
221
222static __inline__ __m128i __DEFAULT_FN_ATTRS
223_mm_rot_epi16(__m128i __A, __m128i __B)
224{
225  return (__m128i)__builtin_ia32_vprotw((__v8hi)__A, (__v8hi)__B);
226}
227
228static __inline__ __m128i __DEFAULT_FN_ATTRS
229_mm_rot_epi32(__m128i __A, __m128i __B)
230{
231  return (__m128i)__builtin_ia32_vprotd((__v4si)__A, (__v4si)__B);
232}
233
234static __inline__ __m128i __DEFAULT_FN_ATTRS
235_mm_rot_epi64(__m128i __A, __m128i __B)
236{
237  return (__m128i)__builtin_ia32_vprotq((__v2di)__A, (__v2di)__B);
238}
239
240#define _mm_roti_epi8(A, N) __extension__ ({ \
241  (__m128i)__builtin_ia32_vprotbi((__v16qi)(__m128i)(A), (N)); })
242
243#define _mm_roti_epi16(A, N) __extension__ ({ \
244  (__m128i)__builtin_ia32_vprotwi((__v8hi)(__m128i)(A), (N)); })
245
246#define _mm_roti_epi32(A, N) __extension__ ({ \
247  (__m128i)__builtin_ia32_vprotdi((__v4si)(__m128i)(A), (N)); })
248
249#define _mm_roti_epi64(A, N) __extension__ ({ \
250  (__m128i)__builtin_ia32_vprotqi((__v2di)(__m128i)(A), (N)); })
251
252static __inline__ __m128i __DEFAULT_FN_ATTRS
253_mm_shl_epi8(__m128i __A, __m128i __B)
254{
255  return (__m128i)__builtin_ia32_vpshlb((__v16qi)__A, (__v16qi)__B);
256}
257
258static __inline__ __m128i __DEFAULT_FN_ATTRS
259_mm_shl_epi16(__m128i __A, __m128i __B)
260{
261  return (__m128i)__builtin_ia32_vpshlw((__v8hi)__A, (__v8hi)__B);
262}
263
264static __inline__ __m128i __DEFAULT_FN_ATTRS
265_mm_shl_epi32(__m128i __A, __m128i __B)
266{
267  return (__m128i)__builtin_ia32_vpshld((__v4si)__A, (__v4si)__B);
268}
269
270static __inline__ __m128i __DEFAULT_FN_ATTRS
271_mm_shl_epi64(__m128i __A, __m128i __B)
272{
273  return (__m128i)__builtin_ia32_vpshlq((__v2di)__A, (__v2di)__B);
274}
275
276static __inline__ __m128i __DEFAULT_FN_ATTRS
277_mm_sha_epi8(__m128i __A, __m128i __B)
278{
279  return (__m128i)__builtin_ia32_vpshab((__v16qi)__A, (__v16qi)__B);
280}
281
282static __inline__ __m128i __DEFAULT_FN_ATTRS
283_mm_sha_epi16(__m128i __A, __m128i __B)
284{
285  return (__m128i)__builtin_ia32_vpshaw((__v8hi)__A, (__v8hi)__B);
286}
287
288static __inline__ __m128i __DEFAULT_FN_ATTRS
289_mm_sha_epi32(__m128i __A, __m128i __B)
290{
291  return (__m128i)__builtin_ia32_vpshad((__v4si)__A, (__v4si)__B);
292}
293
294static __inline__ __m128i __DEFAULT_FN_ATTRS
295_mm_sha_epi64(__m128i __A, __m128i __B)
296{
297  return (__m128i)__builtin_ia32_vpshaq((__v2di)__A, (__v2di)__B);
298}
299
300#define _mm_com_epu8(A, B, N) __extension__ ({ \
301  (__m128i)__builtin_ia32_vpcomub((__v16qi)(__m128i)(A), \
302                                  (__v16qi)(__m128i)(B), (N)); })
303
304#define _mm_com_epu16(A, B, N) __extension__ ({ \
305  (__m128i)__builtin_ia32_vpcomuw((__v8hi)(__m128i)(A), \
306                                  (__v8hi)(__m128i)(B), (N)); })
307
308#define _mm_com_epu32(A, B, N) __extension__ ({ \
309  (__m128i)__builtin_ia32_vpcomud((__v4si)(__m128i)(A), \
310                                  (__v4si)(__m128i)(B), (N)); })
311
312#define _mm_com_epu64(A, B, N) __extension__ ({ \
313  (__m128i)__builtin_ia32_vpcomuq((__v2di)(__m128i)(A), \
314                                  (__v2di)(__m128i)(B), (N)); })
315
316#define _mm_com_epi8(A, B, N) __extension__ ({ \
317  (__m128i)__builtin_ia32_vpcomb((__v16qi)(__m128i)(A), \
318                                 (__v16qi)(__m128i)(B), (N)); })
319
320#define _mm_com_epi16(A, B, N) __extension__ ({ \
321  (__m128i)__builtin_ia32_vpcomw((__v8hi)(__m128i)(A), \
322                                 (__v8hi)(__m128i)(B), (N)); })
323
324#define _mm_com_epi32(A, B, N) __extension__ ({ \
325  (__m128i)__builtin_ia32_vpcomd((__v4si)(__m128i)(A), \
326                                 (__v4si)(__m128i)(B), (N)); })
327
328#define _mm_com_epi64(A, B, N) __extension__ ({ \
329  (__m128i)__builtin_ia32_vpcomq((__v2di)(__m128i)(A), \
330                                 (__v2di)(__m128i)(B), (N)); })
331
332#define _MM_PCOMCTRL_LT    0
333#define _MM_PCOMCTRL_LE    1
334#define _MM_PCOMCTRL_GT    2
335#define _MM_PCOMCTRL_GE    3
336#define _MM_PCOMCTRL_EQ    4
337#define _MM_PCOMCTRL_NEQ   5
338#define _MM_PCOMCTRL_FALSE 6
339#define _MM_PCOMCTRL_TRUE  7
340
341static __inline__ __m128i __DEFAULT_FN_ATTRS
342_mm_comlt_epu8(__m128i __A, __m128i __B)
343{
344  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_LT);
345}
346
347static __inline__ __m128i __DEFAULT_FN_ATTRS
348_mm_comle_epu8(__m128i __A, __m128i __B)
349{
350  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_LE);
351}
352
353static __inline__ __m128i __DEFAULT_FN_ATTRS
354_mm_comgt_epu8(__m128i __A, __m128i __B)
355{
356  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_GT);
357}
358
359static __inline__ __m128i __DEFAULT_FN_ATTRS
360_mm_comge_epu8(__m128i __A, __m128i __B)
361{
362  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_GE);
363}
364
365static __inline__ __m128i __DEFAULT_FN_ATTRS
366_mm_comeq_epu8(__m128i __A, __m128i __B)
367{
368  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_EQ);
369}
370
371static __inline__ __m128i __DEFAULT_FN_ATTRS
372_mm_comneq_epu8(__m128i __A, __m128i __B)
373{
374  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_NEQ);
375}
376
377static __inline__ __m128i __DEFAULT_FN_ATTRS
378_mm_comfalse_epu8(__m128i __A, __m128i __B)
379{
380  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_FALSE);
381}
382
383static __inline__ __m128i __DEFAULT_FN_ATTRS
384_mm_comtrue_epu8(__m128i __A, __m128i __B)
385{
386  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_TRUE);
387}
388
389static __inline__ __m128i __DEFAULT_FN_ATTRS
390_mm_comlt_epu16(__m128i __A, __m128i __B)
391{
392  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_LT);
393}
394
395static __inline__ __m128i __DEFAULT_FN_ATTRS
396_mm_comle_epu16(__m128i __A, __m128i __B)
397{
398  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_LE);
399}
400
401static __inline__ __m128i __DEFAULT_FN_ATTRS
402_mm_comgt_epu16(__m128i __A, __m128i __B)
403{
404  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_GT);
405}
406
407static __inline__ __m128i __DEFAULT_FN_ATTRS
408_mm_comge_epu16(__m128i __A, __m128i __B)
409{
410  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_GE);
411}
412
413static __inline__ __m128i __DEFAULT_FN_ATTRS
414_mm_comeq_epu16(__m128i __A, __m128i __B)
415{
416  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_EQ);
417}
418
419static __inline__ __m128i __DEFAULT_FN_ATTRS
420_mm_comneq_epu16(__m128i __A, __m128i __B)
421{
422  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_NEQ);
423}
424
425static __inline__ __m128i __DEFAULT_FN_ATTRS
426_mm_comfalse_epu16(__m128i __A, __m128i __B)
427{
428  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_FALSE);
429}
430
431static __inline__ __m128i __DEFAULT_FN_ATTRS
432_mm_comtrue_epu16(__m128i __A, __m128i __B)
433{
434  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_TRUE);
435}
436
437static __inline__ __m128i __DEFAULT_FN_ATTRS
438_mm_comlt_epu32(__m128i __A, __m128i __B)
439{
440  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_LT);
441}
442
443static __inline__ __m128i __DEFAULT_FN_ATTRS
444_mm_comle_epu32(__m128i __A, __m128i __B)
445{
446  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_LE);
447}
448
449static __inline__ __m128i __DEFAULT_FN_ATTRS
450_mm_comgt_epu32(__m128i __A, __m128i __B)
451{
452  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_GT);
453}
454
455static __inline__ __m128i __DEFAULT_FN_ATTRS
456_mm_comge_epu32(__m128i __A, __m128i __B)
457{
458  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_GE);
459}
460
461static __inline__ __m128i __DEFAULT_FN_ATTRS
462_mm_comeq_epu32(__m128i __A, __m128i __B)
463{
464  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_EQ);
465}
466
467static __inline__ __m128i __DEFAULT_FN_ATTRS
468_mm_comneq_epu32(__m128i __A, __m128i __B)
469{
470  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_NEQ);
471}
472
473static __inline__ __m128i __DEFAULT_FN_ATTRS
474_mm_comfalse_epu32(__m128i __A, __m128i __B)
475{
476  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_FALSE);
477}
478
479static __inline__ __m128i __DEFAULT_FN_ATTRS
480_mm_comtrue_epu32(__m128i __A, __m128i __B)
481{
482  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_TRUE);
483}
484
485static __inline__ __m128i __DEFAULT_FN_ATTRS
486_mm_comlt_epu64(__m128i __A, __m128i __B)
487{
488  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_LT);
489}
490
491static __inline__ __m128i __DEFAULT_FN_ATTRS
492_mm_comle_epu64(__m128i __A, __m128i __B)
493{
494  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_LE);
495}
496
497static __inline__ __m128i __DEFAULT_FN_ATTRS
498_mm_comgt_epu64(__m128i __A, __m128i __B)
499{
500  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_GT);
501}
502
503static __inline__ __m128i __DEFAULT_FN_ATTRS
504_mm_comge_epu64(__m128i __A, __m128i __B)
505{
506  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_GE);
507}
508
509static __inline__ __m128i __DEFAULT_FN_ATTRS
510_mm_comeq_epu64(__m128i __A, __m128i __B)
511{
512  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_EQ);
513}
514
515static __inline__ __m128i __DEFAULT_FN_ATTRS
516_mm_comneq_epu64(__m128i __A, __m128i __B)
517{
518  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_NEQ);
519}
520
521static __inline__ __m128i __DEFAULT_FN_ATTRS
522_mm_comfalse_epu64(__m128i __A, __m128i __B)
523{
524  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_FALSE);
525}
526
527static __inline__ __m128i __DEFAULT_FN_ATTRS
528_mm_comtrue_epu64(__m128i __A, __m128i __B)
529{
530  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_TRUE);
531}
532
533static __inline__ __m128i __DEFAULT_FN_ATTRS
534_mm_comlt_epi8(__m128i __A, __m128i __B)
535{
536  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_LT);
537}
538
539static __inline__ __m128i __DEFAULT_FN_ATTRS
540_mm_comle_epi8(__m128i __A, __m128i __B)
541{
542  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_LE);
543}
544
545static __inline__ __m128i __DEFAULT_FN_ATTRS
546_mm_comgt_epi8(__m128i __A, __m128i __B)
547{
548  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_GT);
549}
550
551static __inline__ __m128i __DEFAULT_FN_ATTRS
552_mm_comge_epi8(__m128i __A, __m128i __B)
553{
554  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_GE);
555}
556
557static __inline__ __m128i __DEFAULT_FN_ATTRS
558_mm_comeq_epi8(__m128i __A, __m128i __B)
559{
560  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_EQ);
561}
562
563static __inline__ __m128i __DEFAULT_FN_ATTRS
564_mm_comneq_epi8(__m128i __A, __m128i __B)
565{
566  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_NEQ);
567}
568
569static __inline__ __m128i __DEFAULT_FN_ATTRS
570_mm_comfalse_epi8(__m128i __A, __m128i __B)
571{
572  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_FALSE);
573}
574
575static __inline__ __m128i __DEFAULT_FN_ATTRS
576_mm_comtrue_epi8(__m128i __A, __m128i __B)
577{
578  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_TRUE);
579}
580
581static __inline__ __m128i __DEFAULT_FN_ATTRS
582_mm_comlt_epi16(__m128i __A, __m128i __B)
583{
584  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_LT);
585}
586
587static __inline__ __m128i __DEFAULT_FN_ATTRS
588_mm_comle_epi16(__m128i __A, __m128i __B)
589{
590  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_LE);
591}
592
593static __inline__ __m128i __DEFAULT_FN_ATTRS
594_mm_comgt_epi16(__m128i __A, __m128i __B)
595{
596  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_GT);
597}
598
599static __inline__ __m128i __DEFAULT_FN_ATTRS
600_mm_comge_epi16(__m128i __A, __m128i __B)
601{
602  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_GE);
603}
604
605static __inline__ __m128i __DEFAULT_FN_ATTRS
606_mm_comeq_epi16(__m128i __A, __m128i __B)
607{
608  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_EQ);
609}
610
611static __inline__ __m128i __DEFAULT_FN_ATTRS
612_mm_comneq_epi16(__m128i __A, __m128i __B)
613{
614  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_NEQ);
615}
616
617static __inline__ __m128i __DEFAULT_FN_ATTRS
618_mm_comfalse_epi16(__m128i __A, __m128i __B)
619{
620  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_FALSE);
621}
622
623static __inline__ __m128i __DEFAULT_FN_ATTRS
624_mm_comtrue_epi16(__m128i __A, __m128i __B)
625{
626  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_TRUE);
627}
628
629static __inline__ __m128i __DEFAULT_FN_ATTRS
630_mm_comlt_epi32(__m128i __A, __m128i __B)
631{
632  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_LT);
633}
634
635static __inline__ __m128i __DEFAULT_FN_ATTRS
636_mm_comle_epi32(__m128i __A, __m128i __B)
637{
638  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_LE);
639}
640
641static __inline__ __m128i __DEFAULT_FN_ATTRS
642_mm_comgt_epi32(__m128i __A, __m128i __B)
643{
644  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_GT);
645}
646
647static __inline__ __m128i __DEFAULT_FN_ATTRS
648_mm_comge_epi32(__m128i __A, __m128i __B)
649{
650  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_GE);
651}
652
653static __inline__ __m128i __DEFAULT_FN_ATTRS
654_mm_comeq_epi32(__m128i __A, __m128i __B)
655{
656  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_EQ);
657}
658
659static __inline__ __m128i __DEFAULT_FN_ATTRS
660_mm_comneq_epi32(__m128i __A, __m128i __B)
661{
662  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_NEQ);
663}
664
665static __inline__ __m128i __DEFAULT_FN_ATTRS
666_mm_comfalse_epi32(__m128i __A, __m128i __B)
667{
668  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_FALSE);
669}
670
671static __inline__ __m128i __DEFAULT_FN_ATTRS
672_mm_comtrue_epi32(__m128i __A, __m128i __B)
673{
674  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_TRUE);
675}
676
677static __inline__ __m128i __DEFAULT_FN_ATTRS
678_mm_comlt_epi64(__m128i __A, __m128i __B)
679{
680  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_LT);
681}
682
683static __inline__ __m128i __DEFAULT_FN_ATTRS
684_mm_comle_epi64(__m128i __A, __m128i __B)
685{
686  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_LE);
687}
688
689static __inline__ __m128i __DEFAULT_FN_ATTRS
690_mm_comgt_epi64(__m128i __A, __m128i __B)
691{
692  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_GT);
693}
694
695static __inline__ __m128i __DEFAULT_FN_ATTRS
696_mm_comge_epi64(__m128i __A, __m128i __B)
697{
698  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_GE);
699}
700
701static __inline__ __m128i __DEFAULT_FN_ATTRS
702_mm_comeq_epi64(__m128i __A, __m128i __B)
703{
704  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_EQ);
705}
706
707static __inline__ __m128i __DEFAULT_FN_ATTRS
708_mm_comneq_epi64(__m128i __A, __m128i __B)
709{
710  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_NEQ);
711}
712
713static __inline__ __m128i __DEFAULT_FN_ATTRS
714_mm_comfalse_epi64(__m128i __A, __m128i __B)
715{
716  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_FALSE);
717}
718
719static __inline__ __m128i __DEFAULT_FN_ATTRS
720_mm_comtrue_epi64(__m128i __A, __m128i __B)
721{
722  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_TRUE);
723}
724
725#define _mm_permute2_pd(X, Y, C, I) __extension__ ({ \
726  (__m128d)__builtin_ia32_vpermil2pd((__v2df)(__m128d)(X), \
727                                     (__v2df)(__m128d)(Y), \
728                                     (__v2di)(__m128i)(C), (I)); })
729
730#define _mm256_permute2_pd(X, Y, C, I) __extension__ ({ \
731  (__m256d)__builtin_ia32_vpermil2pd256((__v4df)(__m256d)(X), \
732                                        (__v4df)(__m256d)(Y), \
733                                        (__v4di)(__m256i)(C), (I)); })
734
735#define _mm_permute2_ps(X, Y, C, I) __extension__ ({ \
736  (__m128)__builtin_ia32_vpermil2ps((__v4sf)(__m128)(X), (__v4sf)(__m128)(Y), \
737                                    (__v4si)(__m128i)(C), (I)); })
738
739#define _mm256_permute2_ps(X, Y, C, I) __extension__ ({ \
740  (__m256)__builtin_ia32_vpermil2ps256((__v8sf)(__m256)(X), \
741                                       (__v8sf)(__m256)(Y), \
742                                       (__v8si)(__m256i)(C), (I)); })
743
744static __inline__ __m128 __DEFAULT_FN_ATTRS
745_mm_frcz_ss(__m128 __A)
746{
747  return (__m128)__builtin_ia32_vfrczss((__v4sf)__A);
748}
749
750static __inline__ __m128d __DEFAULT_FN_ATTRS
751_mm_frcz_sd(__m128d __A)
752{
753  return (__m128d)__builtin_ia32_vfrczsd((__v2df)__A);
754}
755
756static __inline__ __m128 __DEFAULT_FN_ATTRS
757_mm_frcz_ps(__m128 __A)
758{
759  return (__m128)__builtin_ia32_vfrczps((__v4sf)__A);
760}
761
762static __inline__ __m128d __DEFAULT_FN_ATTRS
763_mm_frcz_pd(__m128d __A)
764{
765  return (__m128d)__builtin_ia32_vfrczpd((__v2df)__A);
766}
767
768static __inline__ __m256 __DEFAULT_FN_ATTRS
769_mm256_frcz_ps(__m256 __A)
770{
771  return (__m256)__builtin_ia32_vfrczps256((__v8sf)__A);
772}
773
774static __inline__ __m256d __DEFAULT_FN_ATTRS
775_mm256_frcz_pd(__m256d __A)
776{
777  return (__m256d)__builtin_ia32_vfrczpd256((__v4df)__A);
778}
779
780#undef __DEFAULT_FN_ATTRS
781
782#endif /* __XOPINTRIN_H */
783