1 /* Copyright (C) 2007-2018 Free Software Foundation, Inc.
2 
3    This file is part of GCC.
4 
5    GCC is free software; you can redistribute it and/or modify
6    it under the terms of the GNU General Public License as published by
7    the Free Software Foundation; either version 3, or (at your option)
8    any later version.
9 
10    GCC is distributed in the hope that it will be useful,
11    but WITHOUT ANY WARRANTY; without even the implied warranty of
12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13    GNU General Public License for more details.
14 
15    Under Section 7 of GPL version 3, you are granted additional
16    permissions described in the GCC Runtime Library Exception, version
17    3.1, as published by the Free Software Foundation.
18 
19    You should have received a copy of the GNU General Public License and
20    a copy of the GCC Runtime Library Exception along with this program;
21    see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
22    <http://www.gnu.org/licenses/>.  */
23 
24 /* Implemented from the specification included in the Intel C++ Compiler
25    User Guide and Reference, version 10.0.  */
26 
27 #ifndef _SMMINTRIN_H_INCLUDED
28 #define _SMMINTRIN_H_INCLUDED
29 
30 /* We need definitions from the SSSE3, SSE3, SSE2 and SSE header
31    files.  */
32 #include <tmmintrin.h>
33 
34 #ifndef __SSE4_1__
35 #pragma GCC push_options
36 #pragma GCC target("sse4.1")
37 #define __DISABLE_SSE4_1__
38 #endif /* __SSE4_1__ */
39 
40 /* Rounding mode macros. */
41 #define _MM_FROUND_TO_NEAREST_INT	0x00
42 #define _MM_FROUND_TO_NEG_INF		0x01
43 #define _MM_FROUND_TO_POS_INF		0x02
44 #define _MM_FROUND_TO_ZERO		0x03
45 #define _MM_FROUND_CUR_DIRECTION	0x04
46 
47 #define _MM_FROUND_RAISE_EXC		0x00
48 #define _MM_FROUND_NO_EXC		0x08
49 
50 #define _MM_FROUND_NINT		\
51   (_MM_FROUND_TO_NEAREST_INT | _MM_FROUND_RAISE_EXC)
52 #define _MM_FROUND_FLOOR	\
53   (_MM_FROUND_TO_NEG_INF | _MM_FROUND_RAISE_EXC)
54 #define _MM_FROUND_CEIL		\
55   (_MM_FROUND_TO_POS_INF | _MM_FROUND_RAISE_EXC)
56 #define _MM_FROUND_TRUNC	\
57   (_MM_FROUND_TO_ZERO | _MM_FROUND_RAISE_EXC)
58 #define _MM_FROUND_RINT		\
59   (_MM_FROUND_CUR_DIRECTION | _MM_FROUND_RAISE_EXC)
60 #define _MM_FROUND_NEARBYINT	\
61   (_MM_FROUND_CUR_DIRECTION | _MM_FROUND_NO_EXC)
62 
63 /* Test Instruction */
64 /* Packed integer 128-bit bitwise comparison. Return 1 if
65    (__V & __M) == 0.  */
66 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
67 _mm_testz_si128 (__m128i __M, __m128i __V)
68 {
69   return __builtin_ia32_ptestz128 ((__v2di)__M, (__v2di)__V);
70 }
71 
72 /* Packed integer 128-bit bitwise comparison. Return 1 if
73    (__V & ~__M) == 0.  */
74 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
75 _mm_testc_si128 (__m128i __M, __m128i __V)
76 {
77   return __builtin_ia32_ptestc128 ((__v2di)__M, (__v2di)__V);
78 }
79 
80 /* Packed integer 128-bit bitwise comparison. Return 1 if
81    (__V & __M) != 0 && (__V & ~__M) != 0.  */
82 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
83 _mm_testnzc_si128 (__m128i __M, __m128i __V)
84 {
85   return __builtin_ia32_ptestnzc128 ((__v2di)__M, (__v2di)__V);
86 }
87 
88 /* Macros for packed integer 128-bit comparison intrinsics.  */
89 #define _mm_test_all_zeros(M, V) _mm_testz_si128 ((M), (V))
90 
91 #define _mm_test_all_ones(V) \
92   _mm_testc_si128 ((V), _mm_cmpeq_epi32 ((V), (V)))
93 
94 #define _mm_test_mix_ones_zeros(M, V) _mm_testnzc_si128 ((M), (V))
95 
96 /* Packed/scalar double precision floating point rounding.  */
97 
98 #ifdef __OPTIMIZE__
99 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
100 _mm_round_pd (__m128d __V, const int __M)
101 {
102   return (__m128d) __builtin_ia32_roundpd ((__v2df)__V, __M);
103 }
104 
105 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
106 _mm_round_sd(__m128d __D, __m128d __V, const int __M)
107 {
108   return (__m128d) __builtin_ia32_roundsd ((__v2df)__D,
109 					   (__v2df)__V,
110 					   __M);
111 }
112 #else
113 #define _mm_round_pd(V, M) \
114   ((__m128d) __builtin_ia32_roundpd ((__v2df)(__m128d)(V), (int)(M)))
115 
116 #define _mm_round_sd(D, V, M)						\
117   ((__m128d) __builtin_ia32_roundsd ((__v2df)(__m128d)(D),		\
118 				     (__v2df)(__m128d)(V), (int)(M)))
119 #endif
120 
121 /* Packed/scalar single precision floating point rounding.  */
122 
123 #ifdef __OPTIMIZE__
124 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
125 _mm_round_ps (__m128 __V, const int __M)
126 {
127   return (__m128) __builtin_ia32_roundps ((__v4sf)__V, __M);
128 }
129 
130 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
131 _mm_round_ss (__m128 __D, __m128 __V, const int __M)
132 {
133   return (__m128) __builtin_ia32_roundss ((__v4sf)__D,
134 					  (__v4sf)__V,
135 					  __M);
136 }
137 #else
138 #define _mm_round_ps(V, M) \
139   ((__m128) __builtin_ia32_roundps ((__v4sf)(__m128)(V), (int)(M)))
140 
141 #define _mm_round_ss(D, V, M)						\
142   ((__m128) __builtin_ia32_roundss ((__v4sf)(__m128)(D),		\
143 				    (__v4sf)(__m128)(V), (int)(M)))
144 #endif
145 
146 /* Macros for ceil/floor intrinsics.  */
147 #define _mm_ceil_pd(V)	   _mm_round_pd ((V), _MM_FROUND_CEIL)
148 #define _mm_ceil_sd(D, V)  _mm_round_sd ((D), (V), _MM_FROUND_CEIL)
149 
150 #define _mm_floor_pd(V)	   _mm_round_pd((V), _MM_FROUND_FLOOR)
151 #define _mm_floor_sd(D, V) _mm_round_sd ((D), (V), _MM_FROUND_FLOOR)
152 
153 #define _mm_ceil_ps(V)	   _mm_round_ps ((V), _MM_FROUND_CEIL)
154 #define _mm_ceil_ss(D, V)  _mm_round_ss ((D), (V), _MM_FROUND_CEIL)
155 
156 #define _mm_floor_ps(V)	   _mm_round_ps ((V), _MM_FROUND_FLOOR)
157 #define _mm_floor_ss(D, V) _mm_round_ss ((D), (V), _MM_FROUND_FLOOR)
158 
159 /* SSE4.1 */
160 
161 /* Integer blend instructions - select data from 2 sources using
162    constant/variable mask.  */
163 
164 #ifdef __OPTIMIZE__
165 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
166 _mm_blend_epi16 (__m128i __X, __m128i __Y, const int __M)
167 {
168   return (__m128i) __builtin_ia32_pblendw128 ((__v8hi)__X,
169 					      (__v8hi)__Y,
170 					      __M);
171 }
172 #else
173 #define _mm_blend_epi16(X, Y, M)					\
174   ((__m128i) __builtin_ia32_pblendw128 ((__v8hi)(__m128i)(X),		\
175 					(__v8hi)(__m128i)(Y), (int)(M)))
176 #endif
177 
178 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
179 _mm_blendv_epi8 (__m128i __X, __m128i __Y, __m128i __M)
180 {
181   return (__m128i) __builtin_ia32_pblendvb128 ((__v16qi)__X,
182 					       (__v16qi)__Y,
183 					       (__v16qi)__M);
184 }
185 
186 /* Single precision floating point blend instructions - select data
187    from 2 sources using constant/variable mask.  */
188 
189 #ifdef __OPTIMIZE__
190 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
191 _mm_blend_ps (__m128 __X, __m128 __Y, const int __M)
192 {
193   return (__m128) __builtin_ia32_blendps ((__v4sf)__X,
194 					  (__v4sf)__Y,
195 					  __M);
196 }
197 #else
198 #define _mm_blend_ps(X, Y, M)						\
199   ((__m128) __builtin_ia32_blendps ((__v4sf)(__m128)(X),		\
200 				    (__v4sf)(__m128)(Y), (int)(M)))
201 #endif
202 
203 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
204 _mm_blendv_ps (__m128 __X, __m128 __Y, __m128 __M)
205 {
206   return (__m128) __builtin_ia32_blendvps ((__v4sf)__X,
207 					   (__v4sf)__Y,
208 					   (__v4sf)__M);
209 }
210 
211 /* Double precision floating point blend instructions - select data
212    from 2 sources using constant/variable mask.  */
213 
214 #ifdef __OPTIMIZE__
215 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
216 _mm_blend_pd (__m128d __X, __m128d __Y, const int __M)
217 {
218   return (__m128d) __builtin_ia32_blendpd ((__v2df)__X,
219 					   (__v2df)__Y,
220 					   __M);
221 }
222 #else
223 #define _mm_blend_pd(X, Y, M)						\
224   ((__m128d) __builtin_ia32_blendpd ((__v2df)(__m128d)(X),		\
225 				     (__v2df)(__m128d)(Y), (int)(M)))
226 #endif
227 
228 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
229 _mm_blendv_pd (__m128d __X, __m128d __Y, __m128d __M)
230 {
231   return (__m128d) __builtin_ia32_blendvpd ((__v2df)__X,
232 					    (__v2df)__Y,
233 					    (__v2df)__M);
234 }
235 
236 /* Dot product instructions with mask-defined summing and zeroing parts
237    of result.  */
238 
239 #ifdef __OPTIMIZE__
240 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
241 _mm_dp_ps (__m128 __X, __m128 __Y, const int __M)
242 {
243   return (__m128) __builtin_ia32_dpps ((__v4sf)__X,
244 				       (__v4sf)__Y,
245 				       __M);
246 }
247 
248 extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
249 _mm_dp_pd (__m128d __X, __m128d __Y, const int __M)
250 {
251   return (__m128d) __builtin_ia32_dppd ((__v2df)__X,
252 					(__v2df)__Y,
253 					__M);
254 }
255 #else
256 #define _mm_dp_ps(X, Y, M)						\
257   ((__m128) __builtin_ia32_dpps ((__v4sf)(__m128)(X),			\
258 				 (__v4sf)(__m128)(Y), (int)(M)))
259 
260 #define _mm_dp_pd(X, Y, M)						\
261   ((__m128d) __builtin_ia32_dppd ((__v2df)(__m128d)(X),			\
262 				  (__v2df)(__m128d)(Y), (int)(M)))
263 #endif
264 
265 /* Packed integer 64-bit comparison, zeroing or filling with ones
266    corresponding parts of result.  */
267 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
268 _mm_cmpeq_epi64 (__m128i __X, __m128i __Y)
269 {
270   return (__m128i) ((__v2di)__X == (__v2di)__Y);
271 }
272 
273 /*  Min/max packed integer instructions.  */
274 
275 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
276 _mm_min_epi8 (__m128i __X, __m128i __Y)
277 {
278   return (__m128i) __builtin_ia32_pminsb128 ((__v16qi)__X, (__v16qi)__Y);
279 }
280 
281 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
282 _mm_max_epi8 (__m128i __X, __m128i __Y)
283 {
284   return (__m128i) __builtin_ia32_pmaxsb128 ((__v16qi)__X, (__v16qi)__Y);
285 }
286 
287 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
288 _mm_min_epu16 (__m128i __X, __m128i __Y)
289 {
290   return (__m128i) __builtin_ia32_pminuw128 ((__v8hi)__X, (__v8hi)__Y);
291 }
292 
293 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
294 _mm_max_epu16 (__m128i __X, __m128i __Y)
295 {
296   return (__m128i) __builtin_ia32_pmaxuw128 ((__v8hi)__X, (__v8hi)__Y);
297 }
298 
299 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
300 _mm_min_epi32 (__m128i __X, __m128i __Y)
301 {
302   return (__m128i) __builtin_ia32_pminsd128 ((__v4si)__X, (__v4si)__Y);
303 }
304 
305 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
306 _mm_max_epi32 (__m128i __X, __m128i __Y)
307 {
308   return (__m128i) __builtin_ia32_pmaxsd128 ((__v4si)__X, (__v4si)__Y);
309 }
310 
311 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
312 _mm_min_epu32 (__m128i __X, __m128i __Y)
313 {
314   return (__m128i) __builtin_ia32_pminud128 ((__v4si)__X, (__v4si)__Y);
315 }
316 
317 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
318 _mm_max_epu32 (__m128i __X, __m128i __Y)
319 {
320   return (__m128i) __builtin_ia32_pmaxud128 ((__v4si)__X, (__v4si)__Y);
321 }
322 
323 /* Packed integer 32-bit multiplication with truncation of upper
324    halves of results.  */
325 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
326 _mm_mullo_epi32 (__m128i __X, __m128i __Y)
327 {
328   return (__m128i) ((__v4su)__X * (__v4su)__Y);
329 }
330 
331 /* Packed integer 32-bit multiplication of 2 pairs of operands
332    with two 64-bit results.  */
333 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
334 _mm_mul_epi32 (__m128i __X, __m128i __Y)
335 {
336   return (__m128i) __builtin_ia32_pmuldq128 ((__v4si)__X, (__v4si)__Y);
337 }
338 
339 /* Insert single precision float into packed single precision array
340    element selected by index N.  The bits [7-6] of N define S
341    index, the bits [5-4] define D index, and bits [3-0] define
342    zeroing mask for D.  */
343 
344 #ifdef __OPTIMIZE__
345 extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
346 _mm_insert_ps (__m128 __D, __m128 __S, const int __N)
347 {
348   return (__m128) __builtin_ia32_insertps128 ((__v4sf)__D,
349 					      (__v4sf)__S,
350 					      __N);
351 }
352 #else
353 #define _mm_insert_ps(D, S, N)						\
354   ((__m128) __builtin_ia32_insertps128 ((__v4sf)(__m128)(D),		\
355 					(__v4sf)(__m128)(S), (int)(N)))
356 #endif
357 
358 /* Helper macro to create the N value for _mm_insert_ps.  */
359 #define _MM_MK_INSERTPS_NDX(S, D, M) (((S) << 6) | ((D) << 4) | (M))
360 
361 /* Extract binary representation of single precision float from packed
362    single precision array element of X selected by index N.  */
363 
364 #ifdef __OPTIMIZE__
365 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
366 _mm_extract_ps (__m128 __X, const int __N)
367 {
368   union { int i; float f; } __tmp;
369   __tmp.f = __builtin_ia32_vec_ext_v4sf ((__v4sf)__X, __N);
370   return __tmp.i;
371 }
372 #else
373 #define _mm_extract_ps(X, N)						\
374   (__extension__							\
375    ({									\
376      union { int i; float f; } __tmp;					\
377      __tmp.f = __builtin_ia32_vec_ext_v4sf ((__v4sf)(__m128)(X), (int)(N)); \
378      __tmp.i;								\
379    }))
380 #endif
381 
382 /* Extract binary representation of single precision float into
383    D from packed single precision array element of S selected
384    by index N.  */
385 #define _MM_EXTRACT_FLOAT(D, S, N) \
386   { (D) = __builtin_ia32_vec_ext_v4sf ((__v4sf)(S), (N)); }
387 
388 /* Extract specified single precision float element into the lower
389    part of __m128.  */
390 #define _MM_PICK_OUT_PS(X, N)				\
391   _mm_insert_ps (_mm_setzero_ps (), (X), 		\
392 		 _MM_MK_INSERTPS_NDX ((N), 0, 0x0e))
393 
394 /* Insert integer, S, into packed integer array element of D
395    selected by index N.  */
396 
397 #ifdef __OPTIMIZE__
398 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
399 _mm_insert_epi8 (__m128i __D, int __S, const int __N)
400 {
401   return (__m128i) __builtin_ia32_vec_set_v16qi ((__v16qi)__D,
402 						 __S, __N);
403 }
404 
405 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
406 _mm_insert_epi32 (__m128i __D, int __S, const int __N)
407 {
408   return (__m128i) __builtin_ia32_vec_set_v4si ((__v4si)__D,
409 						 __S, __N);
410 }
411 
412 #ifdef __x86_64__
413 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
414 _mm_insert_epi64 (__m128i __D, long long __S, const int __N)
415 {
416   return (__m128i) __builtin_ia32_vec_set_v2di ((__v2di)__D,
417 						 __S, __N);
418 }
419 #endif
420 #else
421 #define _mm_insert_epi8(D, S, N)					\
422   ((__m128i) __builtin_ia32_vec_set_v16qi ((__v16qi)(__m128i)(D),	\
423 					   (int)(S), (int)(N)))
424 
425 #define _mm_insert_epi32(D, S, N)				\
426   ((__m128i) __builtin_ia32_vec_set_v4si ((__v4si)(__m128i)(D),	\
427 					  (int)(S), (int)(N)))
428 
429 #ifdef __x86_64__
430 #define _mm_insert_epi64(D, S, N)					\
431   ((__m128i) __builtin_ia32_vec_set_v2di ((__v2di)(__m128i)(D),		\
432 					  (long long)(S), (int)(N)))
433 #endif
434 #endif
435 
436 /* Extract integer from packed integer array element of X selected by
437    index N.  */
438 
439 #ifdef __OPTIMIZE__
440 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
441 _mm_extract_epi8 (__m128i __X, const int __N)
442 {
443    return (unsigned char) __builtin_ia32_vec_ext_v16qi ((__v16qi)__X, __N);
444 }
445 
446 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
447 _mm_extract_epi32 (__m128i __X, const int __N)
448 {
449    return __builtin_ia32_vec_ext_v4si ((__v4si)__X, __N);
450 }
451 
452 #ifdef __x86_64__
453 extern __inline long long  __attribute__((__gnu_inline__, __always_inline__, __artificial__))
454 _mm_extract_epi64 (__m128i __X, const int __N)
455 {
456   return __builtin_ia32_vec_ext_v2di ((__v2di)__X, __N);
457 }
458 #endif
459 #else
460 #define _mm_extract_epi8(X, N) \
461   ((int) (unsigned char) __builtin_ia32_vec_ext_v16qi ((__v16qi)(__m128i)(X), (int)(N)))
462 #define _mm_extract_epi32(X, N) \
463   ((int) __builtin_ia32_vec_ext_v4si ((__v4si)(__m128i)(X), (int)(N)))
464 
465 #ifdef __x86_64__
466 #define _mm_extract_epi64(X, N) \
467   ((long long) __builtin_ia32_vec_ext_v2di ((__v2di)(__m128i)(X), (int)(N)))
468 #endif
469 #endif
470 
471 /* Return horizontal packed word minimum and its index in bits [15:0]
472    and bits [18:16] respectively.  */
473 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
474 _mm_minpos_epu16 (__m128i __X)
475 {
476   return (__m128i) __builtin_ia32_phminposuw128 ((__v8hi)__X);
477 }
478 
479 /* Packed integer sign-extension.  */
480 
481 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
482 _mm_cvtepi8_epi32 (__m128i __X)
483 {
484   return (__m128i) __builtin_ia32_pmovsxbd128 ((__v16qi)__X);
485 }
486 
487 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
488 _mm_cvtepi16_epi32 (__m128i __X)
489 {
490   return (__m128i) __builtin_ia32_pmovsxwd128 ((__v8hi)__X);
491 }
492 
493 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
494 _mm_cvtepi8_epi64 (__m128i __X)
495 {
496   return (__m128i) __builtin_ia32_pmovsxbq128 ((__v16qi)__X);
497 }
498 
499 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
500 _mm_cvtepi32_epi64 (__m128i __X)
501 {
502   return (__m128i) __builtin_ia32_pmovsxdq128 ((__v4si)__X);
503 }
504 
505 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
506 _mm_cvtepi16_epi64 (__m128i __X)
507 {
508   return (__m128i) __builtin_ia32_pmovsxwq128 ((__v8hi)__X);
509 }
510 
511 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
512 _mm_cvtepi8_epi16 (__m128i __X)
513 {
514   return (__m128i) __builtin_ia32_pmovsxbw128 ((__v16qi)__X);
515 }
516 
517 /* Packed integer zero-extension. */
518 
519 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
520 _mm_cvtepu8_epi32 (__m128i __X)
521 {
522   return (__m128i) __builtin_ia32_pmovzxbd128 ((__v16qi)__X);
523 }
524 
525 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
526 _mm_cvtepu16_epi32 (__m128i __X)
527 {
528   return (__m128i) __builtin_ia32_pmovzxwd128 ((__v8hi)__X);
529 }
530 
531 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
532 _mm_cvtepu8_epi64 (__m128i __X)
533 {
534   return (__m128i) __builtin_ia32_pmovzxbq128 ((__v16qi)__X);
535 }
536 
537 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
538 _mm_cvtepu32_epi64 (__m128i __X)
539 {
540   return (__m128i) __builtin_ia32_pmovzxdq128 ((__v4si)__X);
541 }
542 
543 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
544 _mm_cvtepu16_epi64 (__m128i __X)
545 {
546   return (__m128i) __builtin_ia32_pmovzxwq128 ((__v8hi)__X);
547 }
548 
549 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
550 _mm_cvtepu8_epi16 (__m128i __X)
551 {
552   return (__m128i) __builtin_ia32_pmovzxbw128 ((__v16qi)__X);
553 }
554 
555 /* Pack 8 double words from 2 operands into 8 words of result with
556    unsigned saturation. */
557 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
558 _mm_packus_epi32 (__m128i __X, __m128i __Y)
559 {
560   return (__m128i) __builtin_ia32_packusdw128 ((__v4si)__X, (__v4si)__Y);
561 }
562 
563 /* Sum absolute 8-bit integer difference of adjacent groups of 4
564    byte integers in the first 2 operands.  Starting offsets within
565    operands are determined by the 3rd mask operand.  */
566 
567 #ifdef __OPTIMIZE__
568 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
569 _mm_mpsadbw_epu8 (__m128i __X, __m128i __Y, const int __M)
570 {
571   return (__m128i) __builtin_ia32_mpsadbw128 ((__v16qi)__X,
572 					      (__v16qi)__Y, __M);
573 }
574 #else
575 #define _mm_mpsadbw_epu8(X, Y, M)					\
576   ((__m128i) __builtin_ia32_mpsadbw128 ((__v16qi)(__m128i)(X),		\
577 					(__v16qi)(__m128i)(Y), (int)(M)))
578 #endif
579 
580 /* Load double quadword using non-temporal aligned hint.  */
581 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
582 _mm_stream_load_si128 (__m128i *__X)
583 {
584   return (__m128i) __builtin_ia32_movntdqa ((__v2di *) __X);
585 }
586 
587 #ifndef __SSE4_2__
588 #pragma GCC push_options
589 #pragma GCC target("sse4.2")
590 #define __DISABLE_SSE4_2__
591 #endif /* __SSE4_2__ */
592 
593 /* These macros specify the source data format.  */
594 #define _SIDD_UBYTE_OPS			0x00
595 #define _SIDD_UWORD_OPS			0x01
596 #define _SIDD_SBYTE_OPS			0x02
597 #define _SIDD_SWORD_OPS			0x03
598 
599 /* These macros specify the comparison operation.  */
600 #define _SIDD_CMP_EQUAL_ANY		0x00
601 #define _SIDD_CMP_RANGES		0x04
602 #define _SIDD_CMP_EQUAL_EACH		0x08
603 #define _SIDD_CMP_EQUAL_ORDERED		0x0c
604 
605 /* These macros specify the polarity.  */
606 #define _SIDD_POSITIVE_POLARITY		0x00
607 #define _SIDD_NEGATIVE_POLARITY		0x10
608 #define _SIDD_MASKED_POSITIVE_POLARITY	0x20
609 #define _SIDD_MASKED_NEGATIVE_POLARITY	0x30
610 
611 /* These macros specify the output selection in _mm_cmpXstri ().  */
612 #define _SIDD_LEAST_SIGNIFICANT		0x00
613 #define _SIDD_MOST_SIGNIFICANT		0x40
614 
615 /* These macros specify the output selection in _mm_cmpXstrm ().  */
616 #define _SIDD_BIT_MASK			0x00
617 #define _SIDD_UNIT_MASK			0x40
618 
619 /* Intrinsics for text/string processing.  */
620 
621 #ifdef __OPTIMIZE__
622 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
623 _mm_cmpistrm (__m128i __X, __m128i __Y, const int __M)
624 {
625   return (__m128i) __builtin_ia32_pcmpistrm128 ((__v16qi)__X,
626 						(__v16qi)__Y,
627 						__M);
628 }
629 
630 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
631 _mm_cmpistri (__m128i __X, __m128i __Y, const int __M)
632 {
633   return __builtin_ia32_pcmpistri128 ((__v16qi)__X,
634 				      (__v16qi)__Y,
635 				      __M);
636 }
637 
638 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
639 _mm_cmpestrm (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
640 {
641   return (__m128i) __builtin_ia32_pcmpestrm128 ((__v16qi)__X, __LX,
642 						(__v16qi)__Y, __LY,
643 						__M);
644 }
645 
646 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
647 _mm_cmpestri (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
648 {
649   return __builtin_ia32_pcmpestri128 ((__v16qi)__X, __LX,
650 				      (__v16qi)__Y, __LY,
651 				      __M);
652 }
653 #else
654 #define _mm_cmpistrm(X, Y, M)						\
655   ((__m128i) __builtin_ia32_pcmpistrm128 ((__v16qi)(__m128i)(X),	\
656 					  (__v16qi)(__m128i)(Y), (int)(M)))
657 #define _mm_cmpistri(X, Y, M)						\
658   ((int) __builtin_ia32_pcmpistri128 ((__v16qi)(__m128i)(X),		\
659 				      (__v16qi)(__m128i)(Y), (int)(M)))
660 
661 #define _mm_cmpestrm(X, LX, Y, LY, M)					\
662   ((__m128i) __builtin_ia32_pcmpestrm128 ((__v16qi)(__m128i)(X),	\
663 					  (int)(LX), (__v16qi)(__m128i)(Y), \
664 					  (int)(LY), (int)(M)))
665 #define _mm_cmpestri(X, LX, Y, LY, M)					\
666   ((int) __builtin_ia32_pcmpestri128 ((__v16qi)(__m128i)(X), (int)(LX),	\
667 				      (__v16qi)(__m128i)(Y), (int)(LY),	\
668 				      (int)(M)))
669 #endif
670 
671 /* Intrinsics for text/string processing and reading values of
672    EFlags.  */
673 
674 #ifdef __OPTIMIZE__
675 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
676 _mm_cmpistra (__m128i __X, __m128i __Y, const int __M)
677 {
678   return __builtin_ia32_pcmpistria128 ((__v16qi)__X,
679 				       (__v16qi)__Y,
680 				       __M);
681 }
682 
683 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
684 _mm_cmpistrc (__m128i __X, __m128i __Y, const int __M)
685 {
686   return __builtin_ia32_pcmpistric128 ((__v16qi)__X,
687 				       (__v16qi)__Y,
688 				       __M);
689 }
690 
691 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
692 _mm_cmpistro (__m128i __X, __m128i __Y, const int __M)
693 {
694   return __builtin_ia32_pcmpistrio128 ((__v16qi)__X,
695 				       (__v16qi)__Y,
696 				       __M);
697 }
698 
699 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
700 _mm_cmpistrs (__m128i __X, __m128i __Y, const int __M)
701 {
702   return __builtin_ia32_pcmpistris128 ((__v16qi)__X,
703 				       (__v16qi)__Y,
704 				       __M);
705 }
706 
707 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
708 _mm_cmpistrz (__m128i __X, __m128i __Y, const int __M)
709 {
710   return __builtin_ia32_pcmpistriz128 ((__v16qi)__X,
711 				       (__v16qi)__Y,
712 				       __M);
713 }
714 
715 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
716 _mm_cmpestra (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
717 {
718   return __builtin_ia32_pcmpestria128 ((__v16qi)__X, __LX,
719 				       (__v16qi)__Y, __LY,
720 				       __M);
721 }
722 
723 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
724 _mm_cmpestrc (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
725 {
726   return __builtin_ia32_pcmpestric128 ((__v16qi)__X, __LX,
727 				       (__v16qi)__Y, __LY,
728 				       __M);
729 }
730 
731 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
732 _mm_cmpestro (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
733 {
734   return __builtin_ia32_pcmpestrio128 ((__v16qi)__X, __LX,
735 				       (__v16qi)__Y, __LY,
736 				       __M);
737 }
738 
739 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
740 _mm_cmpestrs (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
741 {
742   return __builtin_ia32_pcmpestris128 ((__v16qi)__X, __LX,
743 				       (__v16qi)__Y, __LY,
744 				       __M);
745 }
746 
747 extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
748 _mm_cmpestrz (__m128i __X, int __LX, __m128i __Y, int __LY, const int __M)
749 {
750   return __builtin_ia32_pcmpestriz128 ((__v16qi)__X, __LX,
751 				       (__v16qi)__Y, __LY,
752 				       __M);
753 }
754 #else
755 #define _mm_cmpistra(X, Y, M)						\
756   ((int) __builtin_ia32_pcmpistria128 ((__v16qi)(__m128i)(X),		\
757 				       (__v16qi)(__m128i)(Y), (int)(M)))
758 #define _mm_cmpistrc(X, Y, M)						\
759   ((int) __builtin_ia32_pcmpistric128 ((__v16qi)(__m128i)(X),		\
760 				       (__v16qi)(__m128i)(Y), (int)(M)))
761 #define _mm_cmpistro(X, Y, M)						\
762   ((int) __builtin_ia32_pcmpistrio128 ((__v16qi)(__m128i)(X),		\
763 				       (__v16qi)(__m128i)(Y), (int)(M)))
764 #define _mm_cmpistrs(X, Y, M)						\
765   ((int) __builtin_ia32_pcmpistris128 ((__v16qi)(__m128i)(X),		\
766 				       (__v16qi)(__m128i)(Y), (int)(M)))
767 #define _mm_cmpistrz(X, Y, M)						\
768   ((int) __builtin_ia32_pcmpistriz128 ((__v16qi)(__m128i)(X),		\
769 				       (__v16qi)(__m128i)(Y), (int)(M)))
770 
771 #define _mm_cmpestra(X, LX, Y, LY, M)					\
772   ((int) __builtin_ia32_pcmpestria128 ((__v16qi)(__m128i)(X), (int)(LX), \
773 				       (__v16qi)(__m128i)(Y), (int)(LY), \
774 				       (int)(M)))
775 #define _mm_cmpestrc(X, LX, Y, LY, M)					\
776   ((int) __builtin_ia32_pcmpestric128 ((__v16qi)(__m128i)(X), (int)(LX), \
777 				       (__v16qi)(__m128i)(Y), (int)(LY), \
778 				       (int)(M)))
779 #define _mm_cmpestro(X, LX, Y, LY, M)					\
780   ((int) __builtin_ia32_pcmpestrio128 ((__v16qi)(__m128i)(X), (int)(LX), \
781 				       (__v16qi)(__m128i)(Y), (int)(LY), \
782 				       (int)(M)))
783 #define _mm_cmpestrs(X, LX, Y, LY, M)					\
784   ((int) __builtin_ia32_pcmpestris128 ((__v16qi)(__m128i)(X), (int)(LX), \
785 				       (__v16qi)(__m128i)(Y), (int)(LY), \
786 				       (int)(M)))
787 #define _mm_cmpestrz(X, LX, Y, LY, M)					\
788   ((int) __builtin_ia32_pcmpestriz128 ((__v16qi)(__m128i)(X), (int)(LX), \
789 				       (__v16qi)(__m128i)(Y), (int)(LY), \
790 				       (int)(M)))
791 #endif
792 
793 /* Packed integer 64-bit comparison, zeroing or filling with ones
794    corresponding parts of result.  */
795 extern __inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
796 _mm_cmpgt_epi64 (__m128i __X, __m128i __Y)
797 {
798   return (__m128i) ((__v2di)__X > (__v2di)__Y);
799 }
800 
801 #ifdef __DISABLE_SSE4_2__
802 #undef __DISABLE_SSE4_2__
803 #pragma GCC pop_options
804 #endif /* __DISABLE_SSE4_2__ */
805 
806 #ifdef __DISABLE_SSE4_1__
807 #undef __DISABLE_SSE4_1__
808 #pragma GCC pop_options
809 #endif /* __DISABLE_SSE4_1__ */
810 
811 #include <popcntintrin.h>
812 
813 #ifndef __SSE4_1__
814 #pragma GCC push_options
815 #pragma GCC target("sse4.1")
816 #define __DISABLE_SSE4_1__
817 #endif /* __SSE4_1__ */
818 
819 #ifndef __SSE4_2__
820 #pragma GCC push_options
821 #pragma GCC target("sse4.2")
822 #define __DISABLE_SSE4_2__
823 #endif /* __SSE4_1__ */
824 
825 /* Accumulate CRC32 (polynomial 0x11EDC6F41) value.  */
826 extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
827 _mm_crc32_u8 (unsigned int __C, unsigned char __V)
828 {
829   return __builtin_ia32_crc32qi (__C, __V);
830 }
831 
832 extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
833 _mm_crc32_u16 (unsigned int __C, unsigned short __V)
834 {
835   return __builtin_ia32_crc32hi (__C, __V);
836 }
837 
838 extern __inline unsigned int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
839 _mm_crc32_u32 (unsigned int __C, unsigned int __V)
840 {
841   return __builtin_ia32_crc32si (__C, __V);
842 }
843 
844 #ifdef __x86_64__
845 extern __inline unsigned long long __attribute__((__gnu_inline__, __always_inline__, __artificial__))
846 _mm_crc32_u64 (unsigned long long __C, unsigned long long __V)
847 {
848   return __builtin_ia32_crc32di (__C, __V);
849 }
850 #endif
851 
852 #ifdef __DISABLE_SSE4_2__
853 #undef __DISABLE_SSE4_2__
854 #pragma GCC pop_options
855 #endif /* __DISABLE_SSE4_2__ */
856 
857 #ifdef __DISABLE_SSE4_1__
858 #undef __DISABLE_SSE4_1__
859 #pragma GCC pop_options
860 #endif /* __DISABLE_SSE4_1__ */
861 
862 #endif /* _SMMINTRIN_H_INCLUDED */
863