xopintrin.h revision 288943
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#ifndef __XOP__
32# error "XOP instruction set is not enabled"
33#else
34
35#include <fma4intrin.h>
36
37/* Define the default attributes for the functions in this file. */
38#define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__))
39
40static __inline__ __m128i __DEFAULT_FN_ATTRS
41_mm_maccs_epi16(__m128i __A, __m128i __B, __m128i __C)
42{
43  return (__m128i)__builtin_ia32_vpmacssww((__v8hi)__A, (__v8hi)__B, (__v8hi)__C);
44}
45
46static __inline__ __m128i __DEFAULT_FN_ATTRS
47_mm_macc_epi16(__m128i __A, __m128i __B, __m128i __C)
48{
49  return (__m128i)__builtin_ia32_vpmacsww((__v8hi)__A, (__v8hi)__B, (__v8hi)__C);
50}
51
52static __inline__ __m128i __DEFAULT_FN_ATTRS
53_mm_maccsd_epi16(__m128i __A, __m128i __B, __m128i __C)
54{
55  return (__m128i)__builtin_ia32_vpmacsswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
56}
57
58static __inline__ __m128i __DEFAULT_FN_ATTRS
59_mm_maccd_epi16(__m128i __A, __m128i __B, __m128i __C)
60{
61  return (__m128i)__builtin_ia32_vpmacswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
62}
63
64static __inline__ __m128i __DEFAULT_FN_ATTRS
65_mm_maccs_epi32(__m128i __A, __m128i __B, __m128i __C)
66{
67  return (__m128i)__builtin_ia32_vpmacssdd((__v4si)__A, (__v4si)__B, (__v4si)__C);
68}
69
70static __inline__ __m128i __DEFAULT_FN_ATTRS
71_mm_macc_epi32(__m128i __A, __m128i __B, __m128i __C)
72{
73  return (__m128i)__builtin_ia32_vpmacsdd((__v4si)__A, (__v4si)__B, (__v4si)__C);
74}
75
76static __inline__ __m128i __DEFAULT_FN_ATTRS
77_mm_maccslo_epi32(__m128i __A, __m128i __B, __m128i __C)
78{
79  return (__m128i)__builtin_ia32_vpmacssdql((__v4si)__A, (__v4si)__B, (__v2di)__C);
80}
81
82static __inline__ __m128i __DEFAULT_FN_ATTRS
83_mm_macclo_epi32(__m128i __A, __m128i __B, __m128i __C)
84{
85  return (__m128i)__builtin_ia32_vpmacsdql((__v4si)__A, (__v4si)__B, (__v2di)__C);
86}
87
88static __inline__ __m128i __DEFAULT_FN_ATTRS
89_mm_maccshi_epi32(__m128i __A, __m128i __B, __m128i __C)
90{
91  return (__m128i)__builtin_ia32_vpmacssdqh((__v4si)__A, (__v4si)__B, (__v2di)__C);
92}
93
94static __inline__ __m128i __DEFAULT_FN_ATTRS
95_mm_macchi_epi32(__m128i __A, __m128i __B, __m128i __C)
96{
97  return (__m128i)__builtin_ia32_vpmacsdqh((__v4si)__A, (__v4si)__B, (__v2di)__C);
98}
99
100static __inline__ __m128i __DEFAULT_FN_ATTRS
101_mm_maddsd_epi16(__m128i __A, __m128i __B, __m128i __C)
102{
103  return (__m128i)__builtin_ia32_vpmadcsswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
104}
105
106static __inline__ __m128i __DEFAULT_FN_ATTRS
107_mm_maddd_epi16(__m128i __A, __m128i __B, __m128i __C)
108{
109  return (__m128i)__builtin_ia32_vpmadcswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C);
110}
111
112static __inline__ __m128i __DEFAULT_FN_ATTRS
113_mm_haddw_epi8(__m128i __A)
114{
115  return (__m128i)__builtin_ia32_vphaddbw((__v16qi)__A);
116}
117
118static __inline__ __m128i __DEFAULT_FN_ATTRS
119_mm_haddd_epi8(__m128i __A)
120{
121  return (__m128i)__builtin_ia32_vphaddbd((__v16qi)__A);
122}
123
124static __inline__ __m128i __DEFAULT_FN_ATTRS
125_mm_haddq_epi8(__m128i __A)
126{
127  return (__m128i)__builtin_ia32_vphaddbq((__v16qi)__A);
128}
129
130static __inline__ __m128i __DEFAULT_FN_ATTRS
131_mm_haddd_epi16(__m128i __A)
132{
133  return (__m128i)__builtin_ia32_vphaddwd((__v8hi)__A);
134}
135
136static __inline__ __m128i __DEFAULT_FN_ATTRS
137_mm_haddq_epi16(__m128i __A)
138{
139  return (__m128i)__builtin_ia32_vphaddwq((__v8hi)__A);
140}
141
142static __inline__ __m128i __DEFAULT_FN_ATTRS
143_mm_haddq_epi32(__m128i __A)
144{
145  return (__m128i)__builtin_ia32_vphadddq((__v4si)__A);
146}
147
148static __inline__ __m128i __DEFAULT_FN_ATTRS
149_mm_haddw_epu8(__m128i __A)
150{
151  return (__m128i)__builtin_ia32_vphaddubw((__v16qi)__A);
152}
153
154static __inline__ __m128i __DEFAULT_FN_ATTRS
155_mm_haddd_epu8(__m128i __A)
156{
157  return (__m128i)__builtin_ia32_vphaddubd((__v16qi)__A);
158}
159
160static __inline__ __m128i __DEFAULT_FN_ATTRS
161_mm_haddq_epu8(__m128i __A)
162{
163  return (__m128i)__builtin_ia32_vphaddubq((__v16qi)__A);
164}
165
166static __inline__ __m128i __DEFAULT_FN_ATTRS
167_mm_haddd_epu16(__m128i __A)
168{
169  return (__m128i)__builtin_ia32_vphadduwd((__v8hi)__A);
170}
171
172static __inline__ __m128i __DEFAULT_FN_ATTRS
173_mm_haddq_epu16(__m128i __A)
174{
175  return (__m128i)__builtin_ia32_vphadduwq((__v8hi)__A);
176}
177
178static __inline__ __m128i __DEFAULT_FN_ATTRS
179_mm_haddq_epu32(__m128i __A)
180{
181  return (__m128i)__builtin_ia32_vphaddudq((__v4si)__A);
182}
183
184static __inline__ __m128i __DEFAULT_FN_ATTRS
185_mm_hsubw_epi8(__m128i __A)
186{
187  return (__m128i)__builtin_ia32_vphsubbw((__v16qi)__A);
188}
189
190static __inline__ __m128i __DEFAULT_FN_ATTRS
191_mm_hsubd_epi16(__m128i __A)
192{
193  return (__m128i)__builtin_ia32_vphsubwd((__v8hi)__A);
194}
195
196static __inline__ __m128i __DEFAULT_FN_ATTRS
197_mm_hsubq_epi32(__m128i __A)
198{
199  return (__m128i)__builtin_ia32_vphsubdq((__v4si)__A);
200}
201
202static __inline__ __m128i __DEFAULT_FN_ATTRS
203_mm_cmov_si128(__m128i __A, __m128i __B, __m128i __C)
204{
205  return (__m128i)__builtin_ia32_vpcmov(__A, __B, __C);
206}
207
208static __inline__ __m256i __DEFAULT_FN_ATTRS
209_mm256_cmov_si256(__m256i __A, __m256i __B, __m256i __C)
210{
211  return (__m256i)__builtin_ia32_vpcmov_256(__A, __B, __C);
212}
213
214static __inline__ __m128i __DEFAULT_FN_ATTRS
215_mm_perm_epi8(__m128i __A, __m128i __B, __m128i __C)
216{
217  return (__m128i)__builtin_ia32_vpperm((__v16qi)__A, (__v16qi)__B, (__v16qi)__C);
218}
219
220static __inline__ __m128i __DEFAULT_FN_ATTRS
221_mm_rot_epi8(__m128i __A, __m128i __B)
222{
223  return (__m128i)__builtin_ia32_vprotb((__v16qi)__A, (__v16qi)__B);
224}
225
226static __inline__ __m128i __DEFAULT_FN_ATTRS
227_mm_rot_epi16(__m128i __A, __m128i __B)
228{
229  return (__m128i)__builtin_ia32_vprotw((__v8hi)__A, (__v8hi)__B);
230}
231
232static __inline__ __m128i __DEFAULT_FN_ATTRS
233_mm_rot_epi32(__m128i __A, __m128i __B)
234{
235  return (__m128i)__builtin_ia32_vprotd((__v4si)__A, (__v4si)__B);
236}
237
238static __inline__ __m128i __DEFAULT_FN_ATTRS
239_mm_rot_epi64(__m128i __A, __m128i __B)
240{
241  return (__m128i)__builtin_ia32_vprotq((__v2di)__A, (__v2di)__B);
242}
243
244#define _mm_roti_epi8(A, N) __extension__ ({ \
245  __m128i __A = (A); \
246  (__m128i)__builtin_ia32_vprotbi((__v16qi)__A, (N)); })
247
248#define _mm_roti_epi16(A, N) __extension__ ({ \
249  __m128i __A = (A); \
250  (__m128i)__builtin_ia32_vprotwi((__v8hi)__A, (N)); })
251
252#define _mm_roti_epi32(A, N) __extension__ ({ \
253  __m128i __A = (A); \
254  (__m128i)__builtin_ia32_vprotdi((__v4si)__A, (N)); })
255
256#define _mm_roti_epi64(A, N) __extension__ ({ \
257  __m128i __A = (A); \
258  (__m128i)__builtin_ia32_vprotqi((__v2di)__A, (N)); })
259
260static __inline__ __m128i __DEFAULT_FN_ATTRS
261_mm_shl_epi8(__m128i __A, __m128i __B)
262{
263  return (__m128i)__builtin_ia32_vpshlb((__v16qi)__A, (__v16qi)__B);
264}
265
266static __inline__ __m128i __DEFAULT_FN_ATTRS
267_mm_shl_epi16(__m128i __A, __m128i __B)
268{
269  return (__m128i)__builtin_ia32_vpshlw((__v8hi)__A, (__v8hi)__B);
270}
271
272static __inline__ __m128i __DEFAULT_FN_ATTRS
273_mm_shl_epi32(__m128i __A, __m128i __B)
274{
275  return (__m128i)__builtin_ia32_vpshld((__v4si)__A, (__v4si)__B);
276}
277
278static __inline__ __m128i __DEFAULT_FN_ATTRS
279_mm_shl_epi64(__m128i __A, __m128i __B)
280{
281  return (__m128i)__builtin_ia32_vpshlq((__v2di)__A, (__v2di)__B);
282}
283
284static __inline__ __m128i __DEFAULT_FN_ATTRS
285_mm_sha_epi8(__m128i __A, __m128i __B)
286{
287  return (__m128i)__builtin_ia32_vpshab((__v16qi)__A, (__v16qi)__B);
288}
289
290static __inline__ __m128i __DEFAULT_FN_ATTRS
291_mm_sha_epi16(__m128i __A, __m128i __B)
292{
293  return (__m128i)__builtin_ia32_vpshaw((__v8hi)__A, (__v8hi)__B);
294}
295
296static __inline__ __m128i __DEFAULT_FN_ATTRS
297_mm_sha_epi32(__m128i __A, __m128i __B)
298{
299  return (__m128i)__builtin_ia32_vpshad((__v4si)__A, (__v4si)__B);
300}
301
302static __inline__ __m128i __DEFAULT_FN_ATTRS
303_mm_sha_epi64(__m128i __A, __m128i __B)
304{
305  return (__m128i)__builtin_ia32_vpshaq((__v2di)__A, (__v2di)__B);
306}
307
308#define _mm_com_epu8(A, B, N) __extension__ ({ \
309  __m128i __A = (A); \
310  __m128i __B = (B); \
311  (__m128i)__builtin_ia32_vpcomub((__v16qi)__A, (__v16qi)__B, (N)); })
312
313#define _mm_com_epu16(A, B, N) __extension__ ({ \
314  __m128i __A = (A); \
315  __m128i __B = (B); \
316  (__m128i)__builtin_ia32_vpcomuw((__v8hi)__A, (__v8hi)__B, (N)); })
317
318#define _mm_com_epu32(A, B, N) __extension__ ({ \
319  __m128i __A = (A); \
320  __m128i __B = (B); \
321  (__m128i)__builtin_ia32_vpcomud((__v4si)__A, (__v4si)__B, (N)); })
322
323#define _mm_com_epu64(A, B, N) __extension__ ({ \
324  __m128i __A = (A); \
325  __m128i __B = (B); \
326  (__m128i)__builtin_ia32_vpcomuq((__v2di)__A, (__v2di)__B, (N)); })
327
328#define _mm_com_epi8(A, B, N) __extension__ ({ \
329  __m128i __A = (A); \
330  __m128i __B = (B); \
331  (__m128i)__builtin_ia32_vpcomb((__v16qi)__A, (__v16qi)__B, (N)); })
332
333#define _mm_com_epi16(A, B, N) __extension__ ({ \
334  __m128i __A = (A); \
335  __m128i __B = (B); \
336  (__m128i)__builtin_ia32_vpcomw((__v8hi)__A, (__v8hi)__B, (N)); })
337
338#define _mm_com_epi32(A, B, N) __extension__ ({ \
339  __m128i __A = (A); \
340  __m128i __B = (B); \
341  (__m128i)__builtin_ia32_vpcomd((__v4si)__A, (__v4si)__B, (N)); })
342
343#define _mm_com_epi64(A, B, N) __extension__ ({ \
344  __m128i __A = (A); \
345  __m128i __B = (B); \
346  (__m128i)__builtin_ia32_vpcomq((__v2di)__A, (__v2di)__B, (N)); })
347
348#define _MM_PCOMCTRL_LT    0
349#define _MM_PCOMCTRL_LE    1
350#define _MM_PCOMCTRL_GT    2
351#define _MM_PCOMCTRL_GE    3
352#define _MM_PCOMCTRL_EQ    4
353#define _MM_PCOMCTRL_NEQ   5
354#define _MM_PCOMCTRL_FALSE 6
355#define _MM_PCOMCTRL_TRUE  7
356
357static __inline__ __m128i __DEFAULT_FN_ATTRS
358_mm_comlt_epu8(__m128i __A, __m128i __B)
359{
360  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_LT);
361}
362
363static __inline__ __m128i __DEFAULT_FN_ATTRS
364_mm_comle_epu8(__m128i __A, __m128i __B)
365{
366  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_LE);
367}
368
369static __inline__ __m128i __DEFAULT_FN_ATTRS
370_mm_comgt_epu8(__m128i __A, __m128i __B)
371{
372  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_GT);
373}
374
375static __inline__ __m128i __DEFAULT_FN_ATTRS
376_mm_comge_epu8(__m128i __A, __m128i __B)
377{
378  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_GE);
379}
380
381static __inline__ __m128i __DEFAULT_FN_ATTRS
382_mm_comeq_epu8(__m128i __A, __m128i __B)
383{
384  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_EQ);
385}
386
387static __inline__ __m128i __DEFAULT_FN_ATTRS
388_mm_comneq_epu8(__m128i __A, __m128i __B)
389{
390  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_NEQ);
391}
392
393static __inline__ __m128i __DEFAULT_FN_ATTRS
394_mm_comfalse_epu8(__m128i __A, __m128i __B)
395{
396  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_FALSE);
397}
398
399static __inline__ __m128i __DEFAULT_FN_ATTRS
400_mm_comtrue_epu8(__m128i __A, __m128i __B)
401{
402  return _mm_com_epu8(__A, __B, _MM_PCOMCTRL_TRUE);
403}
404
405static __inline__ __m128i __DEFAULT_FN_ATTRS
406_mm_comlt_epu16(__m128i __A, __m128i __B)
407{
408  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_LT);
409}
410
411static __inline__ __m128i __DEFAULT_FN_ATTRS
412_mm_comle_epu16(__m128i __A, __m128i __B)
413{
414  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_LE);
415}
416
417static __inline__ __m128i __DEFAULT_FN_ATTRS
418_mm_comgt_epu16(__m128i __A, __m128i __B)
419{
420  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_GT);
421}
422
423static __inline__ __m128i __DEFAULT_FN_ATTRS
424_mm_comge_epu16(__m128i __A, __m128i __B)
425{
426  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_GE);
427}
428
429static __inline__ __m128i __DEFAULT_FN_ATTRS
430_mm_comeq_epu16(__m128i __A, __m128i __B)
431{
432  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_EQ);
433}
434
435static __inline__ __m128i __DEFAULT_FN_ATTRS
436_mm_comneq_epu16(__m128i __A, __m128i __B)
437{
438  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_NEQ);
439}
440
441static __inline__ __m128i __DEFAULT_FN_ATTRS
442_mm_comfalse_epu16(__m128i __A, __m128i __B)
443{
444  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_FALSE);
445}
446
447static __inline__ __m128i __DEFAULT_FN_ATTRS
448_mm_comtrue_epu16(__m128i __A, __m128i __B)
449{
450  return _mm_com_epu16(__A, __B, _MM_PCOMCTRL_TRUE);
451}
452
453static __inline__ __m128i __DEFAULT_FN_ATTRS
454_mm_comlt_epu32(__m128i __A, __m128i __B)
455{
456  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_LT);
457}
458
459static __inline__ __m128i __DEFAULT_FN_ATTRS
460_mm_comle_epu32(__m128i __A, __m128i __B)
461{
462  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_LE);
463}
464
465static __inline__ __m128i __DEFAULT_FN_ATTRS
466_mm_comgt_epu32(__m128i __A, __m128i __B)
467{
468  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_GT);
469}
470
471static __inline__ __m128i __DEFAULT_FN_ATTRS
472_mm_comge_epu32(__m128i __A, __m128i __B)
473{
474  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_GE);
475}
476
477static __inline__ __m128i __DEFAULT_FN_ATTRS
478_mm_comeq_epu32(__m128i __A, __m128i __B)
479{
480  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_EQ);
481}
482
483static __inline__ __m128i __DEFAULT_FN_ATTRS
484_mm_comneq_epu32(__m128i __A, __m128i __B)
485{
486  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_NEQ);
487}
488
489static __inline__ __m128i __DEFAULT_FN_ATTRS
490_mm_comfalse_epu32(__m128i __A, __m128i __B)
491{
492  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_FALSE);
493}
494
495static __inline__ __m128i __DEFAULT_FN_ATTRS
496_mm_comtrue_epu32(__m128i __A, __m128i __B)
497{
498  return _mm_com_epu32(__A, __B, _MM_PCOMCTRL_TRUE);
499}
500
501static __inline__ __m128i __DEFAULT_FN_ATTRS
502_mm_comlt_epu64(__m128i __A, __m128i __B)
503{
504  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_LT);
505}
506
507static __inline__ __m128i __DEFAULT_FN_ATTRS
508_mm_comle_epu64(__m128i __A, __m128i __B)
509{
510  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_LE);
511}
512
513static __inline__ __m128i __DEFAULT_FN_ATTRS
514_mm_comgt_epu64(__m128i __A, __m128i __B)
515{
516  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_GT);
517}
518
519static __inline__ __m128i __DEFAULT_FN_ATTRS
520_mm_comge_epu64(__m128i __A, __m128i __B)
521{
522  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_GE);
523}
524
525static __inline__ __m128i __DEFAULT_FN_ATTRS
526_mm_comeq_epu64(__m128i __A, __m128i __B)
527{
528  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_EQ);
529}
530
531static __inline__ __m128i __DEFAULT_FN_ATTRS
532_mm_comneq_epu64(__m128i __A, __m128i __B)
533{
534  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_NEQ);
535}
536
537static __inline__ __m128i __DEFAULT_FN_ATTRS
538_mm_comfalse_epu64(__m128i __A, __m128i __B)
539{
540  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_FALSE);
541}
542
543static __inline__ __m128i __DEFAULT_FN_ATTRS
544_mm_comtrue_epu64(__m128i __A, __m128i __B)
545{
546  return _mm_com_epu64(__A, __B, _MM_PCOMCTRL_TRUE);
547}
548
549static __inline__ __m128i __DEFAULT_FN_ATTRS
550_mm_comlt_epi8(__m128i __A, __m128i __B)
551{
552  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_LT);
553}
554
555static __inline__ __m128i __DEFAULT_FN_ATTRS
556_mm_comle_epi8(__m128i __A, __m128i __B)
557{
558  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_LE);
559}
560
561static __inline__ __m128i __DEFAULT_FN_ATTRS
562_mm_comgt_epi8(__m128i __A, __m128i __B)
563{
564  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_GT);
565}
566
567static __inline__ __m128i __DEFAULT_FN_ATTRS
568_mm_comge_epi8(__m128i __A, __m128i __B)
569{
570  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_GE);
571}
572
573static __inline__ __m128i __DEFAULT_FN_ATTRS
574_mm_comeq_epi8(__m128i __A, __m128i __B)
575{
576  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_EQ);
577}
578
579static __inline__ __m128i __DEFAULT_FN_ATTRS
580_mm_comneq_epi8(__m128i __A, __m128i __B)
581{
582  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_NEQ);
583}
584
585static __inline__ __m128i __DEFAULT_FN_ATTRS
586_mm_comfalse_epi8(__m128i __A, __m128i __B)
587{
588  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_FALSE);
589}
590
591static __inline__ __m128i __DEFAULT_FN_ATTRS
592_mm_comtrue_epi8(__m128i __A, __m128i __B)
593{
594  return _mm_com_epi8(__A, __B, _MM_PCOMCTRL_TRUE);
595}
596
597static __inline__ __m128i __DEFAULT_FN_ATTRS
598_mm_comlt_epi16(__m128i __A, __m128i __B)
599{
600  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_LT);
601}
602
603static __inline__ __m128i __DEFAULT_FN_ATTRS
604_mm_comle_epi16(__m128i __A, __m128i __B)
605{
606  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_LE);
607}
608
609static __inline__ __m128i __DEFAULT_FN_ATTRS
610_mm_comgt_epi16(__m128i __A, __m128i __B)
611{
612  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_GT);
613}
614
615static __inline__ __m128i __DEFAULT_FN_ATTRS
616_mm_comge_epi16(__m128i __A, __m128i __B)
617{
618  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_GE);
619}
620
621static __inline__ __m128i __DEFAULT_FN_ATTRS
622_mm_comeq_epi16(__m128i __A, __m128i __B)
623{
624  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_EQ);
625}
626
627static __inline__ __m128i __DEFAULT_FN_ATTRS
628_mm_comneq_epi16(__m128i __A, __m128i __B)
629{
630  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_NEQ);
631}
632
633static __inline__ __m128i __DEFAULT_FN_ATTRS
634_mm_comfalse_epi16(__m128i __A, __m128i __B)
635{
636  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_FALSE);
637}
638
639static __inline__ __m128i __DEFAULT_FN_ATTRS
640_mm_comtrue_epi16(__m128i __A, __m128i __B)
641{
642  return _mm_com_epi16(__A, __B, _MM_PCOMCTRL_TRUE);
643}
644
645static __inline__ __m128i __DEFAULT_FN_ATTRS
646_mm_comlt_epi32(__m128i __A, __m128i __B)
647{
648  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_LT);
649}
650
651static __inline__ __m128i __DEFAULT_FN_ATTRS
652_mm_comle_epi32(__m128i __A, __m128i __B)
653{
654  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_LE);
655}
656
657static __inline__ __m128i __DEFAULT_FN_ATTRS
658_mm_comgt_epi32(__m128i __A, __m128i __B)
659{
660  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_GT);
661}
662
663static __inline__ __m128i __DEFAULT_FN_ATTRS
664_mm_comge_epi32(__m128i __A, __m128i __B)
665{
666  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_GE);
667}
668
669static __inline__ __m128i __DEFAULT_FN_ATTRS
670_mm_comeq_epi32(__m128i __A, __m128i __B)
671{
672  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_EQ);
673}
674
675static __inline__ __m128i __DEFAULT_FN_ATTRS
676_mm_comneq_epi32(__m128i __A, __m128i __B)
677{
678  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_NEQ);
679}
680
681static __inline__ __m128i __DEFAULT_FN_ATTRS
682_mm_comfalse_epi32(__m128i __A, __m128i __B)
683{
684  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_FALSE);
685}
686
687static __inline__ __m128i __DEFAULT_FN_ATTRS
688_mm_comtrue_epi32(__m128i __A, __m128i __B)
689{
690  return _mm_com_epi32(__A, __B, _MM_PCOMCTRL_TRUE);
691}
692
693static __inline__ __m128i __DEFAULT_FN_ATTRS
694_mm_comlt_epi64(__m128i __A, __m128i __B)
695{
696  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_LT);
697}
698
699static __inline__ __m128i __DEFAULT_FN_ATTRS
700_mm_comle_epi64(__m128i __A, __m128i __B)
701{
702  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_LE);
703}
704
705static __inline__ __m128i __DEFAULT_FN_ATTRS
706_mm_comgt_epi64(__m128i __A, __m128i __B)
707{
708  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_GT);
709}
710
711static __inline__ __m128i __DEFAULT_FN_ATTRS
712_mm_comge_epi64(__m128i __A, __m128i __B)
713{
714  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_GE);
715}
716
717static __inline__ __m128i __DEFAULT_FN_ATTRS
718_mm_comeq_epi64(__m128i __A, __m128i __B)
719{
720  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_EQ);
721}
722
723static __inline__ __m128i __DEFAULT_FN_ATTRS
724_mm_comneq_epi64(__m128i __A, __m128i __B)
725{
726  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_NEQ);
727}
728
729static __inline__ __m128i __DEFAULT_FN_ATTRS
730_mm_comfalse_epi64(__m128i __A, __m128i __B)
731{
732  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_FALSE);
733}
734
735static __inline__ __m128i __DEFAULT_FN_ATTRS
736_mm_comtrue_epi64(__m128i __A, __m128i __B)
737{
738  return _mm_com_epi64(__A, __B, _MM_PCOMCTRL_TRUE);
739}
740
741#define _mm_permute2_pd(X, Y, C, I) __extension__ ({ \
742  __m128d __X = (X); \
743  __m128d __Y = (Y); \
744  __m128i __C = (C); \
745  (__m128d)__builtin_ia32_vpermil2pd((__v2df)__X, (__v2df)__Y, \
746                                     (__v2di)__C, (I)); })
747
748#define _mm256_permute2_pd(X, Y, C, I) __extension__ ({ \
749  __m256d __X = (X); \
750  __m256d __Y = (Y); \
751  __m256i __C = (C); \
752  (__m256d)__builtin_ia32_vpermil2pd256((__v4df)__X, (__v4df)__Y, \
753                                        (__v4di)__C, (I)); })
754
755#define _mm_permute2_ps(X, Y, C, I) __extension__ ({ \
756  __m128 __X = (X); \
757  __m128 __Y = (Y); \
758  __m128i __C = (C); \
759  (__m128)__builtin_ia32_vpermil2ps((__v4sf)__X, (__v4sf)__Y, \
760                                    (__v4si)__C, (I)); })
761
762#define _mm256_permute2_ps(X, Y, C, I) __extension__ ({ \
763  __m256 __X = (X); \
764  __m256 __Y = (Y); \
765  __m256i __C = (C); \
766  (__m256)__builtin_ia32_vpermil2ps256((__v8sf)__X, (__v8sf)__Y, \
767                                       (__v8si)__C, (I)); })
768
769static __inline__ __m128 __DEFAULT_FN_ATTRS
770_mm_frcz_ss(__m128 __A)
771{
772  return (__m128)__builtin_ia32_vfrczss((__v4sf)__A);
773}
774
775static __inline__ __m128d __DEFAULT_FN_ATTRS
776_mm_frcz_sd(__m128d __A)
777{
778  return (__m128d)__builtin_ia32_vfrczsd((__v2df)__A);
779}
780
781static __inline__ __m128 __DEFAULT_FN_ATTRS
782_mm_frcz_ps(__m128 __A)
783{
784  return (__m128)__builtin_ia32_vfrczps((__v4sf)__A);
785}
786
787static __inline__ __m128d __DEFAULT_FN_ATTRS
788_mm_frcz_pd(__m128d __A)
789{
790  return (__m128d)__builtin_ia32_vfrczpd((__v2df)__A);
791}
792
793static __inline__ __m256 __DEFAULT_FN_ATTRS
794_mm256_frcz_ps(__m256 __A)
795{
796  return (__m256)__builtin_ia32_vfrczps256((__v8sf)__A);
797}
798
799static __inline__ __m256d __DEFAULT_FN_ATTRS
800_mm256_frcz_pd(__m256d __A)
801{
802  return (__m256d)__builtin_ia32_vfrczpd256((__v4df)__A);
803}
804
805#undef __DEFAULT_FN_ATTRS
806
807#endif /* __XOP__ */
808
809#endif /* __XOPINTRIN_H */
810