1 /*===---- mmintrin.h - MMX 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 *===-----------------------------------------------------------------------===
28 #error "MMX instruction set not enabled"
31 typedef long long __m64 __attribute__((__vector_size__(8)));
33 typedef int __v2si __attribute__((__vector_size__(8)));
34 typedef short __v4hi __attribute__((__vector_size__(8)));
35 typedef char __v8qi __attribute__((__vector_size__(8)));
37 static __inline__ void __attribute__((__always_inline__, __nodebug__))
40 __builtin_ia32_emms();
43 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
44 _mm_cvtsi32_si64(int __i)
46 return (__m64)__builtin_ia32_vec_init_v2si(__i, 0);
49 static __inline__ int __attribute__((__always_inline__, __nodebug__))
50 _mm_cvtsi64_si32(__m64 __m)
52 return __builtin_ia32_vec_ext_v2si((__v2si)__m, 0);
55 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
56 _mm_cvtsi64_m64(long long __i)
61 static __inline__ long long __attribute__((__always_inline__, __nodebug__))
62 _mm_cvtm64_si64(__m64 __m)
64 return (long long)__m;
67 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
68 _mm_packs_pi16(__m64 __m1, __m64 __m2)
70 return (__m64)__builtin_ia32_packsswb((__v4hi)__m1, (__v4hi)__m2);
73 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
74 _mm_packs_pi32(__m64 __m1, __m64 __m2)
76 return (__m64)__builtin_ia32_packssdw((__v2si)__m1, (__v2si)__m2);
79 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
80 _mm_packs_pu16(__m64 __m1, __m64 __m2)
82 return (__m64)__builtin_ia32_packuswb((__v4hi)__m1, (__v4hi)__m2);
85 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
86 _mm_unpackhi_pi8(__m64 __m1, __m64 __m2)
88 return (__m64)__builtin_ia32_punpckhbw((__v8qi)__m1, (__v8qi)__m2);
91 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
92 _mm_unpackhi_pi16(__m64 __m1, __m64 __m2)
94 return (__m64)__builtin_ia32_punpckhwd((__v4hi)__m1, (__v4hi)__m2);
97 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
98 _mm_unpackhi_pi32(__m64 __m1, __m64 __m2)
100 return (__m64)__builtin_ia32_punpckhdq((__v2si)__m1, (__v2si)__m2);
103 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
104 _mm_unpacklo_pi8(__m64 __m1, __m64 __m2)
106 return (__m64)__builtin_ia32_punpcklbw((__v8qi)__m1, (__v8qi)__m2);
109 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
110 _mm_unpacklo_pi16(__m64 __m1, __m64 __m2)
112 return (__m64)__builtin_ia32_punpcklwd((__v4hi)__m1, (__v4hi)__m2);
115 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
116 _mm_unpacklo_pi32(__m64 __m1, __m64 __m2)
118 return (__m64)__builtin_ia32_punpckldq((__v2si)__m1, (__v2si)__m2);
121 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
122 _mm_add_pi8(__m64 __m1, __m64 __m2)
124 return (__m64)__builtin_ia32_paddb((__v8qi)__m1, (__v8qi)__m2);
127 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
128 _mm_add_pi16(__m64 __m1, __m64 __m2)
130 return (__m64)__builtin_ia32_paddw((__v4hi)__m1, (__v4hi)__m2);
133 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
134 _mm_add_pi32(__m64 __m1, __m64 __m2)
136 return (__m64)__builtin_ia32_paddd((__v2si)__m1, (__v2si)__m2);
139 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
140 _mm_adds_pi8(__m64 __m1, __m64 __m2)
142 return (__m64)__builtin_ia32_paddsb((__v8qi)__m1, (__v8qi)__m2);
145 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
146 _mm_adds_pi16(__m64 __m1, __m64 __m2)
148 return (__m64)__builtin_ia32_paddsw((__v4hi)__m1, (__v4hi)__m2);
151 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
152 _mm_adds_pu8(__m64 __m1, __m64 __m2)
154 return (__m64)__builtin_ia32_paddusb((__v8qi)__m1, (__v8qi)__m2);
157 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
158 _mm_adds_pu16(__m64 __m1, __m64 __m2)
160 return (__m64)__builtin_ia32_paddusw((__v4hi)__m1, (__v4hi)__m2);
163 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
164 _mm_sub_pi8(__m64 __m1, __m64 __m2)
166 return (__m64)__builtin_ia32_psubb((__v8qi)__m1, (__v8qi)__m2);
169 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
170 _mm_sub_pi16(__m64 __m1, __m64 __m2)
172 return (__m64)__builtin_ia32_psubw((__v4hi)__m1, (__v4hi)__m2);
175 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
176 _mm_sub_pi32(__m64 __m1, __m64 __m2)
178 return (__m64)__builtin_ia32_psubd((__v2si)__m1, (__v2si)__m2);
181 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
182 _mm_subs_pi8(__m64 __m1, __m64 __m2)
184 return (__m64)__builtin_ia32_psubsb((__v8qi)__m1, (__v8qi)__m2);
187 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
188 _mm_subs_pi16(__m64 __m1, __m64 __m2)
190 return (__m64)__builtin_ia32_psubsw((__v4hi)__m1, (__v4hi)__m2);
193 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
194 _mm_subs_pu8(__m64 __m1, __m64 __m2)
196 return (__m64)__builtin_ia32_psubusb((__v8qi)__m1, (__v8qi)__m2);
199 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
200 _mm_subs_pu16(__m64 __m1, __m64 __m2)
202 return (__m64)__builtin_ia32_psubusw((__v4hi)__m1, (__v4hi)__m2);
205 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
206 _mm_madd_pi16(__m64 __m1, __m64 __m2)
208 return (__m64)__builtin_ia32_pmaddwd((__v4hi)__m1, (__v4hi)__m2);
211 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
212 _mm_mulhi_pi16(__m64 __m1, __m64 __m2)
214 return (__m64)__builtin_ia32_pmulhw((__v4hi)__m1, (__v4hi)__m2);
217 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
218 _mm_mullo_pi16(__m64 __m1, __m64 __m2)
220 return (__m64)__builtin_ia32_pmullw((__v4hi)__m1, (__v4hi)__m2);
223 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
224 _mm_sll_pi16(__m64 __m, __m64 __count)
226 return (__m64)__builtin_ia32_psllw((__v4hi)__m, __count);
229 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
230 _mm_slli_pi16(__m64 __m, int __count)
232 return (__m64)__builtin_ia32_psllwi((__v4hi)__m, __count);
235 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
236 _mm_sll_pi32(__m64 __m, __m64 __count)
238 return (__m64)__builtin_ia32_pslld((__v2si)__m, __count);
241 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
242 _mm_slli_pi32(__m64 __m, int __count)
244 return (__m64)__builtin_ia32_pslldi((__v2si)__m, __count);
247 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
248 _mm_sll_si64(__m64 __m, __m64 __count)
250 return (__m64)__builtin_ia32_psllq(__m, __count);
253 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
254 _mm_slli_si64(__m64 __m, int __count)
256 return (__m64)__builtin_ia32_psllqi(__m, __count);
259 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
260 _mm_sra_pi16(__m64 __m, __m64 __count)
262 return (__m64)__builtin_ia32_psraw((__v4hi)__m, __count);
265 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
266 _mm_srai_pi16(__m64 __m, int __count)
268 return (__m64)__builtin_ia32_psrawi((__v4hi)__m, __count);
271 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
272 _mm_sra_pi32(__m64 __m, __m64 __count)
274 return (__m64)__builtin_ia32_psrad((__v2si)__m, __count);
277 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
278 _mm_srai_pi32(__m64 __m, int __count)
280 return (__m64)__builtin_ia32_psradi((__v2si)__m, __count);
283 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
284 _mm_srl_pi16(__m64 __m, __m64 __count)
286 return (__m64)__builtin_ia32_psrlw((__v4hi)__m, __count);
289 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
290 _mm_srli_pi16(__m64 __m, int __count)
292 return (__m64)__builtin_ia32_psrlwi((__v4hi)__m, __count);
295 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
296 _mm_srl_pi32(__m64 __m, __m64 __count)
298 return (__m64)__builtin_ia32_psrld((__v2si)__m, __count);
301 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
302 _mm_srli_pi32(__m64 __m, int __count)
304 return (__m64)__builtin_ia32_psrldi((__v2si)__m, __count);
307 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
308 _mm_srl_si64(__m64 __m, __m64 __count)
310 return (__m64)__builtin_ia32_psrlq(__m, __count);
313 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
314 _mm_srli_si64(__m64 __m, int __count)
316 return (__m64)__builtin_ia32_psrlqi(__m, __count);
319 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
320 _mm_and_si64(__m64 __m1, __m64 __m2)
322 return __builtin_ia32_pand(__m1, __m2);
325 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
326 _mm_andnot_si64(__m64 __m1, __m64 __m2)
328 return __builtin_ia32_pandn(__m1, __m2);
331 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
332 _mm_or_si64(__m64 __m1, __m64 __m2)
334 return __builtin_ia32_por(__m1, __m2);
337 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
338 _mm_xor_si64(__m64 __m1, __m64 __m2)
340 return __builtin_ia32_pxor(__m1, __m2);
343 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
344 _mm_cmpeq_pi8(__m64 __m1, __m64 __m2)
346 return (__m64)__builtin_ia32_pcmpeqb((__v8qi)__m1, (__v8qi)__m2);
349 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
350 _mm_cmpeq_pi16(__m64 __m1, __m64 __m2)
352 return (__m64)__builtin_ia32_pcmpeqw((__v4hi)__m1, (__v4hi)__m2);
355 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
356 _mm_cmpeq_pi32(__m64 __m1, __m64 __m2)
358 return (__m64)__builtin_ia32_pcmpeqd((__v2si)__m1, (__v2si)__m2);
361 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
362 _mm_cmpgt_pi8(__m64 __m1, __m64 __m2)
364 return (__m64)__builtin_ia32_pcmpgtb((__v8qi)__m1, (__v8qi)__m2);
367 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
368 _mm_cmpgt_pi16(__m64 __m1, __m64 __m2)
370 return (__m64)__builtin_ia32_pcmpgtw((__v4hi)__m1, (__v4hi)__m2);
373 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
374 _mm_cmpgt_pi32(__m64 __m1, __m64 __m2)
376 return (__m64)__builtin_ia32_pcmpgtd((__v2si)__m1, (__v2si)__m2);
379 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
380 _mm_setzero_si64(void)
382 return (__m64){ 0LL };
385 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
386 _mm_set_pi32(int __i1, int __i0)
388 return (__m64)__builtin_ia32_vec_init_v2si(__i0, __i1);
391 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
392 _mm_set_pi16(short __s3, short __s2, short __s1, short __s0)
394 return (__m64)__builtin_ia32_vec_init_v4hi(__s0, __s1, __s2, __s3);
397 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
398 _mm_set_pi8(char __b7, char __b6, char __b5, char __b4, char __b3, char __b2,
399 char __b1, char __b0)
401 return (__m64)__builtin_ia32_vec_init_v8qi(__b0, __b1, __b2, __b3,
402 __b4, __b5, __b6, __b7);
405 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
406 _mm_set1_pi32(int __i)
408 return _mm_set_pi32(__i, __i);
411 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
412 _mm_set1_pi16(short __w)
414 return _mm_set_pi16(__w, __w, __w, __w);
417 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
418 _mm_set1_pi8(char __b)
420 return _mm_set_pi8(__b, __b, __b, __b, __b, __b, __b, __b);
423 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
424 _mm_setr_pi32(int __i0, int __i1)
426 return _mm_set_pi32(__i1, __i0);
429 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
430 _mm_setr_pi16(short __w0, short __w1, short __w2, short __w3)
432 return _mm_set_pi16(__w3, __w2, __w1, __w0);
435 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
436 _mm_setr_pi8(char __b0, char __b1, char __b2, char __b3, char __b4, char __b5,
437 char __b6, char __b7)
439 return _mm_set_pi8(__b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0);
443 /* Aliases for compatibility. */
444 #define _m_empty _mm_empty
445 #define _m_from_int _mm_cvtsi32_si64
446 #define _m_to_int _mm_cvtsi64_si32
447 #define _m_packsswb _mm_packs_pi16
448 #define _m_packssdw _mm_packs_pi32
449 #define _m_packuswb _mm_packs_pu16
450 #define _m_punpckhbw _mm_unpackhi_pi8
451 #define _m_punpckhwd _mm_unpackhi_pi16
452 #define _m_punpckhdq _mm_unpackhi_pi32
453 #define _m_punpcklbw _mm_unpacklo_pi8
454 #define _m_punpcklwd _mm_unpacklo_pi16
455 #define _m_punpckldq _mm_unpacklo_pi32
456 #define _m_paddb _mm_add_pi8
457 #define _m_paddw _mm_add_pi16
458 #define _m_paddd _mm_add_pi32
459 #define _m_paddsb _mm_adds_pi8
460 #define _m_paddsw _mm_adds_pi16
461 #define _m_paddusb _mm_adds_pu8
462 #define _m_paddusw _mm_adds_pu16
463 #define _m_psubb _mm_sub_pi8
464 #define _m_psubw _mm_sub_pi16
465 #define _m_psubd _mm_sub_pi32
466 #define _m_psubsb _mm_subs_pi8
467 #define _m_psubsw _mm_subs_pi16
468 #define _m_psubusb _mm_subs_pu8
469 #define _m_psubusw _mm_subs_pu16
470 #define _m_pmaddwd _mm_madd_pi16
471 #define _m_pmulhw _mm_mulhi_pi16
472 #define _m_pmullw _mm_mullo_pi16
473 #define _m_psllw _mm_sll_pi16
474 #define _m_psllwi _mm_slli_pi16
475 #define _m_pslld _mm_sll_pi32
476 #define _m_pslldi _mm_slli_pi32
477 #define _m_psllq _mm_sll_si64
478 #define _m_psllqi _mm_slli_si64
479 #define _m_psraw _mm_sra_pi16
480 #define _m_psrawi _mm_srai_pi16
481 #define _m_psrad _mm_sra_pi32
482 #define _m_psradi _mm_srai_pi32
483 #define _m_psrlw _mm_srl_pi16
484 #define _m_psrlwi _mm_srli_pi16
485 #define _m_psrld _mm_srl_pi32
486 #define _m_psrldi _mm_srli_pi32
487 #define _m_psrlq _mm_srl_si64
488 #define _m_psrlqi _mm_srli_si64
489 #define _m_pand _mm_and_si64
490 #define _m_pandn _mm_andnot_si64
491 #define _m_por _mm_or_si64
492 #define _m_pxor _mm_xor_si64
493 #define _m_pcmpeqb _mm_cmpeq_pi8
494 #define _m_pcmpeqw _mm_cmpeq_pi16
495 #define _m_pcmpeqd _mm_cmpeq_pi32
496 #define _m_pcmpgtb _mm_cmpgt_pi8
497 #define _m_pcmpgtw _mm_cmpgt_pi16
498 #define _m_pcmpgtd _mm_cmpgt_pi32
502 #endif /* __MMINTRIN_H */