1 /*===---- xmmintrin.h - SSE 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 __XMMINTRIN_H
25 #define __XMMINTRIN_H
26
27 #include <mmintrin.h>
28
29 typedef int __v4si __attribute__((__vector_size__(16)));
30 typedef float __v4sf __attribute__((__vector_size__(16)));
31 typedef float __m128 __attribute__((__vector_size__(16)));
32
33 /* This header should only be included in a hosted environment as it depends on
34 * a standard library to provide allocation routines. */
35 #if __STDC_HOSTED__
36 #include <mm_malloc.h>
37 #endif
38
39 /* Define the default attributes for the functions in this file. */
40 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("sse")))
41
42 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_add_ss(__m128 __a,__m128 __b)43 _mm_add_ss(__m128 __a, __m128 __b)
44 {
45 __a[0] += __b[0];
46 return __a;
47 }
48
49 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_add_ps(__m128 __a,__m128 __b)50 _mm_add_ps(__m128 __a, __m128 __b)
51 {
52 return __a + __b;
53 }
54
55 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_sub_ss(__m128 __a,__m128 __b)56 _mm_sub_ss(__m128 __a, __m128 __b)
57 {
58 __a[0] -= __b[0];
59 return __a;
60 }
61
62 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_sub_ps(__m128 __a,__m128 __b)63 _mm_sub_ps(__m128 __a, __m128 __b)
64 {
65 return __a - __b;
66 }
67
68 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_mul_ss(__m128 __a,__m128 __b)69 _mm_mul_ss(__m128 __a, __m128 __b)
70 {
71 __a[0] *= __b[0];
72 return __a;
73 }
74
75 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_mul_ps(__m128 __a,__m128 __b)76 _mm_mul_ps(__m128 __a, __m128 __b)
77 {
78 return __a * __b;
79 }
80
81 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_div_ss(__m128 __a,__m128 __b)82 _mm_div_ss(__m128 __a, __m128 __b)
83 {
84 __a[0] /= __b[0];
85 return __a;
86 }
87
88 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_div_ps(__m128 __a,__m128 __b)89 _mm_div_ps(__m128 __a, __m128 __b)
90 {
91 return __a / __b;
92 }
93
94 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_sqrt_ss(__m128 __a)95 _mm_sqrt_ss(__m128 __a)
96 {
97 __m128 __c = __builtin_ia32_sqrtss(__a);
98 return (__m128) { __c[0], __a[1], __a[2], __a[3] };
99 }
100
101 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_sqrt_ps(__m128 __a)102 _mm_sqrt_ps(__m128 __a)
103 {
104 return __builtin_ia32_sqrtps(__a);
105 }
106
107 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_rcp_ss(__m128 __a)108 _mm_rcp_ss(__m128 __a)
109 {
110 __m128 __c = __builtin_ia32_rcpss(__a);
111 return (__m128) { __c[0], __a[1], __a[2], __a[3] };
112 }
113
114 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_rcp_ps(__m128 __a)115 _mm_rcp_ps(__m128 __a)
116 {
117 return __builtin_ia32_rcpps(__a);
118 }
119
120 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_rsqrt_ss(__m128 __a)121 _mm_rsqrt_ss(__m128 __a)
122 {
123 __m128 __c = __builtin_ia32_rsqrtss(__a);
124 return (__m128) { __c[0], __a[1], __a[2], __a[3] };
125 }
126
127 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_rsqrt_ps(__m128 __a)128 _mm_rsqrt_ps(__m128 __a)
129 {
130 return __builtin_ia32_rsqrtps(__a);
131 }
132
133 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_min_ss(__m128 __a,__m128 __b)134 _mm_min_ss(__m128 __a, __m128 __b)
135 {
136 return __builtin_ia32_minss(__a, __b);
137 }
138
139 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_min_ps(__m128 __a,__m128 __b)140 _mm_min_ps(__m128 __a, __m128 __b)
141 {
142 return __builtin_ia32_minps(__a, __b);
143 }
144
145 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_max_ss(__m128 __a,__m128 __b)146 _mm_max_ss(__m128 __a, __m128 __b)
147 {
148 return __builtin_ia32_maxss(__a, __b);
149 }
150
151 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_max_ps(__m128 __a,__m128 __b)152 _mm_max_ps(__m128 __a, __m128 __b)
153 {
154 return __builtin_ia32_maxps(__a, __b);
155 }
156
157 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_and_ps(__m128 __a,__m128 __b)158 _mm_and_ps(__m128 __a, __m128 __b)
159 {
160 return (__m128)((__v4si)__a & (__v4si)__b);
161 }
162
163 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_andnot_ps(__m128 __a,__m128 __b)164 _mm_andnot_ps(__m128 __a, __m128 __b)
165 {
166 return (__m128)(~(__v4si)__a & (__v4si)__b);
167 }
168
169 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_or_ps(__m128 __a,__m128 __b)170 _mm_or_ps(__m128 __a, __m128 __b)
171 {
172 return (__m128)((__v4si)__a | (__v4si)__b);
173 }
174
175 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_xor_ps(__m128 __a,__m128 __b)176 _mm_xor_ps(__m128 __a, __m128 __b)
177 {
178 return (__m128)((__v4si)__a ^ (__v4si)__b);
179 }
180
181 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpeq_ss(__m128 __a,__m128 __b)182 _mm_cmpeq_ss(__m128 __a, __m128 __b)
183 {
184 return (__m128)__builtin_ia32_cmpeqss(__a, __b);
185 }
186
187 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpeq_ps(__m128 __a,__m128 __b)188 _mm_cmpeq_ps(__m128 __a, __m128 __b)
189 {
190 return (__m128)__builtin_ia32_cmpeqps(__a, __b);
191 }
192
193 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmplt_ss(__m128 __a,__m128 __b)194 _mm_cmplt_ss(__m128 __a, __m128 __b)
195 {
196 return (__m128)__builtin_ia32_cmpltss(__a, __b);
197 }
198
199 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmplt_ps(__m128 __a,__m128 __b)200 _mm_cmplt_ps(__m128 __a, __m128 __b)
201 {
202 return (__m128)__builtin_ia32_cmpltps(__a, __b);
203 }
204
205 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmple_ss(__m128 __a,__m128 __b)206 _mm_cmple_ss(__m128 __a, __m128 __b)
207 {
208 return (__m128)__builtin_ia32_cmpless(__a, __b);
209 }
210
211 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmple_ps(__m128 __a,__m128 __b)212 _mm_cmple_ps(__m128 __a, __m128 __b)
213 {
214 return (__m128)__builtin_ia32_cmpleps(__a, __b);
215 }
216
217 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpgt_ss(__m128 __a,__m128 __b)218 _mm_cmpgt_ss(__m128 __a, __m128 __b)
219 {
220 return (__m128)__builtin_shufflevector(__a,
221 __builtin_ia32_cmpltss(__b, __a),
222 4, 1, 2, 3);
223 }
224
225 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpgt_ps(__m128 __a,__m128 __b)226 _mm_cmpgt_ps(__m128 __a, __m128 __b)
227 {
228 return (__m128)__builtin_ia32_cmpltps(__b, __a);
229 }
230
231 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpge_ss(__m128 __a,__m128 __b)232 _mm_cmpge_ss(__m128 __a, __m128 __b)
233 {
234 return (__m128)__builtin_shufflevector(__a,
235 __builtin_ia32_cmpless(__b, __a),
236 4, 1, 2, 3);
237 }
238
239 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpge_ps(__m128 __a,__m128 __b)240 _mm_cmpge_ps(__m128 __a, __m128 __b)
241 {
242 return (__m128)__builtin_ia32_cmpleps(__b, __a);
243 }
244
245 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpneq_ss(__m128 __a,__m128 __b)246 _mm_cmpneq_ss(__m128 __a, __m128 __b)
247 {
248 return (__m128)__builtin_ia32_cmpneqss(__a, __b);
249 }
250
251 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpneq_ps(__m128 __a,__m128 __b)252 _mm_cmpneq_ps(__m128 __a, __m128 __b)
253 {
254 return (__m128)__builtin_ia32_cmpneqps(__a, __b);
255 }
256
257 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpnlt_ss(__m128 __a,__m128 __b)258 _mm_cmpnlt_ss(__m128 __a, __m128 __b)
259 {
260 return (__m128)__builtin_ia32_cmpnltss(__a, __b);
261 }
262
263 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpnlt_ps(__m128 __a,__m128 __b)264 _mm_cmpnlt_ps(__m128 __a, __m128 __b)
265 {
266 return (__m128)__builtin_ia32_cmpnltps(__a, __b);
267 }
268
269 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpnle_ss(__m128 __a,__m128 __b)270 _mm_cmpnle_ss(__m128 __a, __m128 __b)
271 {
272 return (__m128)__builtin_ia32_cmpnless(__a, __b);
273 }
274
275 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpnle_ps(__m128 __a,__m128 __b)276 _mm_cmpnle_ps(__m128 __a, __m128 __b)
277 {
278 return (__m128)__builtin_ia32_cmpnleps(__a, __b);
279 }
280
281 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpngt_ss(__m128 __a,__m128 __b)282 _mm_cmpngt_ss(__m128 __a, __m128 __b)
283 {
284 return (__m128)__builtin_shufflevector(__a,
285 __builtin_ia32_cmpnltss(__b, __a),
286 4, 1, 2, 3);
287 }
288
289 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpngt_ps(__m128 __a,__m128 __b)290 _mm_cmpngt_ps(__m128 __a, __m128 __b)
291 {
292 return (__m128)__builtin_ia32_cmpnltps(__b, __a);
293 }
294
295 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpnge_ss(__m128 __a,__m128 __b)296 _mm_cmpnge_ss(__m128 __a, __m128 __b)
297 {
298 return (__m128)__builtin_shufflevector(__a,
299 __builtin_ia32_cmpnless(__b, __a),
300 4, 1, 2, 3);
301 }
302
303 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpnge_ps(__m128 __a,__m128 __b)304 _mm_cmpnge_ps(__m128 __a, __m128 __b)
305 {
306 return (__m128)__builtin_ia32_cmpnleps(__b, __a);
307 }
308
309 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpord_ss(__m128 __a,__m128 __b)310 _mm_cmpord_ss(__m128 __a, __m128 __b)
311 {
312 return (__m128)__builtin_ia32_cmpordss(__a, __b);
313 }
314
315 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpord_ps(__m128 __a,__m128 __b)316 _mm_cmpord_ps(__m128 __a, __m128 __b)
317 {
318 return (__m128)__builtin_ia32_cmpordps(__a, __b);
319 }
320
321 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpunord_ss(__m128 __a,__m128 __b)322 _mm_cmpunord_ss(__m128 __a, __m128 __b)
323 {
324 return (__m128)__builtin_ia32_cmpunordss(__a, __b);
325 }
326
327 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cmpunord_ps(__m128 __a,__m128 __b)328 _mm_cmpunord_ps(__m128 __a, __m128 __b)
329 {
330 return (__m128)__builtin_ia32_cmpunordps(__a, __b);
331 }
332
333 static __inline__ int __DEFAULT_FN_ATTRS
_mm_comieq_ss(__m128 __a,__m128 __b)334 _mm_comieq_ss(__m128 __a, __m128 __b)
335 {
336 return __builtin_ia32_comieq(__a, __b);
337 }
338
339 static __inline__ int __DEFAULT_FN_ATTRS
_mm_comilt_ss(__m128 __a,__m128 __b)340 _mm_comilt_ss(__m128 __a, __m128 __b)
341 {
342 return __builtin_ia32_comilt(__a, __b);
343 }
344
345 static __inline__ int __DEFAULT_FN_ATTRS
_mm_comile_ss(__m128 __a,__m128 __b)346 _mm_comile_ss(__m128 __a, __m128 __b)
347 {
348 return __builtin_ia32_comile(__a, __b);
349 }
350
351 static __inline__ int __DEFAULT_FN_ATTRS
_mm_comigt_ss(__m128 __a,__m128 __b)352 _mm_comigt_ss(__m128 __a, __m128 __b)
353 {
354 return __builtin_ia32_comigt(__a, __b);
355 }
356
357 static __inline__ int __DEFAULT_FN_ATTRS
_mm_comige_ss(__m128 __a,__m128 __b)358 _mm_comige_ss(__m128 __a, __m128 __b)
359 {
360 return __builtin_ia32_comige(__a, __b);
361 }
362
363 static __inline__ int __DEFAULT_FN_ATTRS
_mm_comineq_ss(__m128 __a,__m128 __b)364 _mm_comineq_ss(__m128 __a, __m128 __b)
365 {
366 return __builtin_ia32_comineq(__a, __b);
367 }
368
369 static __inline__ int __DEFAULT_FN_ATTRS
_mm_ucomieq_ss(__m128 __a,__m128 __b)370 _mm_ucomieq_ss(__m128 __a, __m128 __b)
371 {
372 return __builtin_ia32_ucomieq(__a, __b);
373 }
374
375 static __inline__ int __DEFAULT_FN_ATTRS
_mm_ucomilt_ss(__m128 __a,__m128 __b)376 _mm_ucomilt_ss(__m128 __a, __m128 __b)
377 {
378 return __builtin_ia32_ucomilt(__a, __b);
379 }
380
381 static __inline__ int __DEFAULT_FN_ATTRS
_mm_ucomile_ss(__m128 __a,__m128 __b)382 _mm_ucomile_ss(__m128 __a, __m128 __b)
383 {
384 return __builtin_ia32_ucomile(__a, __b);
385 }
386
387 static __inline__ int __DEFAULT_FN_ATTRS
_mm_ucomigt_ss(__m128 __a,__m128 __b)388 _mm_ucomigt_ss(__m128 __a, __m128 __b)
389 {
390 return __builtin_ia32_ucomigt(__a, __b);
391 }
392
393 static __inline__ int __DEFAULT_FN_ATTRS
_mm_ucomige_ss(__m128 __a,__m128 __b)394 _mm_ucomige_ss(__m128 __a, __m128 __b)
395 {
396 return __builtin_ia32_ucomige(__a, __b);
397 }
398
399 static __inline__ int __DEFAULT_FN_ATTRS
_mm_ucomineq_ss(__m128 __a,__m128 __b)400 _mm_ucomineq_ss(__m128 __a, __m128 __b)
401 {
402 return __builtin_ia32_ucomineq(__a, __b);
403 }
404
405 static __inline__ int __DEFAULT_FN_ATTRS
_mm_cvtss_si32(__m128 __a)406 _mm_cvtss_si32(__m128 __a)
407 {
408 return __builtin_ia32_cvtss2si(__a);
409 }
410
411 static __inline__ int __DEFAULT_FN_ATTRS
_mm_cvt_ss2si(__m128 __a)412 _mm_cvt_ss2si(__m128 __a)
413 {
414 return _mm_cvtss_si32(__a);
415 }
416
417 #ifdef __x86_64__
418
419 static __inline__ long long __DEFAULT_FN_ATTRS
_mm_cvtss_si64(__m128 __a)420 _mm_cvtss_si64(__m128 __a)
421 {
422 return __builtin_ia32_cvtss2si64(__a);
423 }
424
425 #endif
426
427 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_cvtps_pi32(__m128 __a)428 _mm_cvtps_pi32(__m128 __a)
429 {
430 return (__m64)__builtin_ia32_cvtps2pi(__a);
431 }
432
433 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_cvt_ps2pi(__m128 __a)434 _mm_cvt_ps2pi(__m128 __a)
435 {
436 return _mm_cvtps_pi32(__a);
437 }
438
439 static __inline__ int __DEFAULT_FN_ATTRS
_mm_cvttss_si32(__m128 __a)440 _mm_cvttss_si32(__m128 __a)
441 {
442 return __a[0];
443 }
444
445 static __inline__ int __DEFAULT_FN_ATTRS
_mm_cvtt_ss2si(__m128 __a)446 _mm_cvtt_ss2si(__m128 __a)
447 {
448 return _mm_cvttss_si32(__a);
449 }
450
451 static __inline__ long long __DEFAULT_FN_ATTRS
_mm_cvttss_si64(__m128 __a)452 _mm_cvttss_si64(__m128 __a)
453 {
454 return __a[0];
455 }
456
457 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_cvttps_pi32(__m128 __a)458 _mm_cvttps_pi32(__m128 __a)
459 {
460 return (__m64)__builtin_ia32_cvttps2pi(__a);
461 }
462
463 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_cvtt_ps2pi(__m128 __a)464 _mm_cvtt_ps2pi(__m128 __a)
465 {
466 return _mm_cvttps_pi32(__a);
467 }
468
469 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cvtsi32_ss(__m128 __a,int __b)470 _mm_cvtsi32_ss(__m128 __a, int __b)
471 {
472 __a[0] = __b;
473 return __a;
474 }
475
476 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cvt_si2ss(__m128 __a,int __b)477 _mm_cvt_si2ss(__m128 __a, int __b)
478 {
479 return _mm_cvtsi32_ss(__a, __b);
480 }
481
482 #ifdef __x86_64__
483
484 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cvtsi64_ss(__m128 __a,long long __b)485 _mm_cvtsi64_ss(__m128 __a, long long __b)
486 {
487 __a[0] = __b;
488 return __a;
489 }
490
491 #endif
492
493 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cvtpi32_ps(__m128 __a,__m64 __b)494 _mm_cvtpi32_ps(__m128 __a, __m64 __b)
495 {
496 return __builtin_ia32_cvtpi2ps(__a, (__v2si)__b);
497 }
498
499 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cvt_pi2ps(__m128 __a,__m64 __b)500 _mm_cvt_pi2ps(__m128 __a, __m64 __b)
501 {
502 return _mm_cvtpi32_ps(__a, __b);
503 }
504
505 static __inline__ float __DEFAULT_FN_ATTRS
_mm_cvtss_f32(__m128 __a)506 _mm_cvtss_f32(__m128 __a)
507 {
508 return __a[0];
509 }
510
511 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_loadh_pi(__m128 __a,const __m64 * __p)512 _mm_loadh_pi(__m128 __a, const __m64 *__p)
513 {
514 typedef float __mm_loadh_pi_v2f32 __attribute__((__vector_size__(8)));
515 struct __mm_loadh_pi_struct {
516 __mm_loadh_pi_v2f32 __u;
517 } __attribute__((__packed__, __may_alias__));
518 __mm_loadh_pi_v2f32 __b = ((struct __mm_loadh_pi_struct*)__p)->__u;
519 __m128 __bb = __builtin_shufflevector(__b, __b, 0, 1, 0, 1);
520 return __builtin_shufflevector(__a, __bb, 0, 1, 4, 5);
521 }
522
523 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_loadl_pi(__m128 __a,const __m64 * __p)524 _mm_loadl_pi(__m128 __a, const __m64 *__p)
525 {
526 typedef float __mm_loadl_pi_v2f32 __attribute__((__vector_size__(8)));
527 struct __mm_loadl_pi_struct {
528 __mm_loadl_pi_v2f32 __u;
529 } __attribute__((__packed__, __may_alias__));
530 __mm_loadl_pi_v2f32 __b = ((struct __mm_loadl_pi_struct*)__p)->__u;
531 __m128 __bb = __builtin_shufflevector(__b, __b, 0, 1, 0, 1);
532 return __builtin_shufflevector(__a, __bb, 4, 5, 2, 3);
533 }
534
535 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_load_ss(const float * __p)536 _mm_load_ss(const float *__p)
537 {
538 struct __mm_load_ss_struct {
539 float __u;
540 } __attribute__((__packed__, __may_alias__));
541 float __u = ((struct __mm_load_ss_struct*)__p)->__u;
542 return (__m128){ __u, 0, 0, 0 };
543 }
544
545 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_load1_ps(const float * __p)546 _mm_load1_ps(const float *__p)
547 {
548 struct __mm_load1_ps_struct {
549 float __u;
550 } __attribute__((__packed__, __may_alias__));
551 float __u = ((struct __mm_load1_ps_struct*)__p)->__u;
552 return (__m128){ __u, __u, __u, __u };
553 }
554
555 #define _mm_load_ps1(p) _mm_load1_ps(p)
556
557 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_load_ps(const float * __p)558 _mm_load_ps(const float *__p)
559 {
560 return *(__m128*)__p;
561 }
562
563 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_loadu_ps(const float * __p)564 _mm_loadu_ps(const float *__p)
565 {
566 struct __loadu_ps {
567 __m128 __v;
568 } __attribute__((__packed__, __may_alias__));
569 return ((struct __loadu_ps*)__p)->__v;
570 }
571
572 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_loadr_ps(const float * __p)573 _mm_loadr_ps(const float *__p)
574 {
575 __m128 __a = _mm_load_ps(__p);
576 return __builtin_shufflevector(__a, __a, 3, 2, 1, 0);
577 }
578
579 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_undefined_ps()580 _mm_undefined_ps()
581 {
582 return (__m128)__builtin_ia32_undef128();
583 }
584
585 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_set_ss(float __w)586 _mm_set_ss(float __w)
587 {
588 return (__m128){ __w, 0, 0, 0 };
589 }
590
591 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_set1_ps(float __w)592 _mm_set1_ps(float __w)
593 {
594 return (__m128){ __w, __w, __w, __w };
595 }
596
597 /* Microsoft specific. */
598 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_set_ps1(float __w)599 _mm_set_ps1(float __w)
600 {
601 return _mm_set1_ps(__w);
602 }
603
604 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_set_ps(float __z,float __y,float __x,float __w)605 _mm_set_ps(float __z, float __y, float __x, float __w)
606 {
607 return (__m128){ __w, __x, __y, __z };
608 }
609
610 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_setr_ps(float __z,float __y,float __x,float __w)611 _mm_setr_ps(float __z, float __y, float __x, float __w)
612 {
613 return (__m128){ __z, __y, __x, __w };
614 }
615
616 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_setzero_ps(void)617 _mm_setzero_ps(void)
618 {
619 return (__m128){ 0, 0, 0, 0 };
620 }
621
622 static __inline__ void __DEFAULT_FN_ATTRS
_mm_storeh_pi(__m64 * __p,__m128 __a)623 _mm_storeh_pi(__m64 *__p, __m128 __a)
624 {
625 __builtin_ia32_storehps((__v2si *)__p, __a);
626 }
627
628 static __inline__ void __DEFAULT_FN_ATTRS
_mm_storel_pi(__m64 * __p,__m128 __a)629 _mm_storel_pi(__m64 *__p, __m128 __a)
630 {
631 __builtin_ia32_storelps((__v2si *)__p, __a);
632 }
633
634 static __inline__ void __DEFAULT_FN_ATTRS
_mm_store_ss(float * __p,__m128 __a)635 _mm_store_ss(float *__p, __m128 __a)
636 {
637 struct __mm_store_ss_struct {
638 float __u;
639 } __attribute__((__packed__, __may_alias__));
640 ((struct __mm_store_ss_struct*)__p)->__u = __a[0];
641 }
642
643 static __inline__ void __DEFAULT_FN_ATTRS
_mm_storeu_ps(float * __p,__m128 __a)644 _mm_storeu_ps(float *__p, __m128 __a)
645 {
646 __builtin_ia32_storeups(__p, __a);
647 }
648
649 static __inline__ void __DEFAULT_FN_ATTRS
_mm_store1_ps(float * __p,__m128 __a)650 _mm_store1_ps(float *__p, __m128 __a)
651 {
652 __a = __builtin_shufflevector(__a, __a, 0, 0, 0, 0);
653 _mm_storeu_ps(__p, __a);
654 }
655
656 static __inline__ void __DEFAULT_FN_ATTRS
_mm_store_ps1(float * __p,__m128 __a)657 _mm_store_ps1(float *__p, __m128 __a)
658 {
659 return _mm_store1_ps(__p, __a);
660 }
661
662 static __inline__ void __DEFAULT_FN_ATTRS
_mm_store_ps(float * __p,__m128 __a)663 _mm_store_ps(float *__p, __m128 __a)
664 {
665 *(__m128 *)__p = __a;
666 }
667
668 static __inline__ void __DEFAULT_FN_ATTRS
_mm_storer_ps(float * __p,__m128 __a)669 _mm_storer_ps(float *__p, __m128 __a)
670 {
671 __a = __builtin_shufflevector(__a, __a, 3, 2, 1, 0);
672 _mm_store_ps(__p, __a);
673 }
674
675 #define _MM_HINT_T0 3
676 #define _MM_HINT_T1 2
677 #define _MM_HINT_T2 1
678 #define _MM_HINT_NTA 0
679
680 #ifndef _MSC_VER
681 /* FIXME: We have to #define this because "sel" must be a constant integer, and
682 Sema doesn't do any form of constant propagation yet. */
683
684 #define _mm_prefetch(a, sel) (__builtin_prefetch((void *)(a), 0, (sel)))
685 #endif
686
687 static __inline__ void __DEFAULT_FN_ATTRS
_mm_stream_pi(__m64 * __p,__m64 __a)688 _mm_stream_pi(__m64 *__p, __m64 __a)
689 {
690 __builtin_ia32_movntq(__p, __a);
691 }
692
693 static __inline__ void __DEFAULT_FN_ATTRS
_mm_stream_ps(float * __p,__m128 __a)694 _mm_stream_ps(float *__p, __m128 __a)
695 {
696 __builtin_ia32_movntps(__p, __a);
697 }
698
699 static __inline__ void __DEFAULT_FN_ATTRS
_mm_sfence(void)700 _mm_sfence(void)
701 {
702 __builtin_ia32_sfence();
703 }
704
705 static __inline__ int __DEFAULT_FN_ATTRS
_mm_extract_pi16(__m64 __a,int __n)706 _mm_extract_pi16(__m64 __a, int __n)
707 {
708 __v4hi __b = (__v4hi)__a;
709 return (unsigned short)__b[__n & 3];
710 }
711
712 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_insert_pi16(__m64 __a,int __d,int __n)713 _mm_insert_pi16(__m64 __a, int __d, int __n)
714 {
715 __v4hi __b = (__v4hi)__a;
716 __b[__n & 3] = __d;
717 return (__m64)__b;
718 }
719
720 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_max_pi16(__m64 __a,__m64 __b)721 _mm_max_pi16(__m64 __a, __m64 __b)
722 {
723 return (__m64)__builtin_ia32_pmaxsw((__v4hi)__a, (__v4hi)__b);
724 }
725
726 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_max_pu8(__m64 __a,__m64 __b)727 _mm_max_pu8(__m64 __a, __m64 __b)
728 {
729 return (__m64)__builtin_ia32_pmaxub((__v8qi)__a, (__v8qi)__b);
730 }
731
732 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_min_pi16(__m64 __a,__m64 __b)733 _mm_min_pi16(__m64 __a, __m64 __b)
734 {
735 return (__m64)__builtin_ia32_pminsw((__v4hi)__a, (__v4hi)__b);
736 }
737
738 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_min_pu8(__m64 __a,__m64 __b)739 _mm_min_pu8(__m64 __a, __m64 __b)
740 {
741 return (__m64)__builtin_ia32_pminub((__v8qi)__a, (__v8qi)__b);
742 }
743
744 static __inline__ int __DEFAULT_FN_ATTRS
_mm_movemask_pi8(__m64 __a)745 _mm_movemask_pi8(__m64 __a)
746 {
747 return __builtin_ia32_pmovmskb((__v8qi)__a);
748 }
749
750 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_mulhi_pu16(__m64 __a,__m64 __b)751 _mm_mulhi_pu16(__m64 __a, __m64 __b)
752 {
753 return (__m64)__builtin_ia32_pmulhuw((__v4hi)__a, (__v4hi)__b);
754 }
755
756 #define _mm_shuffle_pi16(a, n) __extension__ ({ \
757 (__m64)__builtin_ia32_pshufw((__v4hi)(__m64)(a), (n)); })
758
759 static __inline__ void __DEFAULT_FN_ATTRS
_mm_maskmove_si64(__m64 __d,__m64 __n,char * __p)760 _mm_maskmove_si64(__m64 __d, __m64 __n, char *__p)
761 {
762 __builtin_ia32_maskmovq((__v8qi)__d, (__v8qi)__n, __p);
763 }
764
765 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_avg_pu8(__m64 __a,__m64 __b)766 _mm_avg_pu8(__m64 __a, __m64 __b)
767 {
768 return (__m64)__builtin_ia32_pavgb((__v8qi)__a, (__v8qi)__b);
769 }
770
771 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_avg_pu16(__m64 __a,__m64 __b)772 _mm_avg_pu16(__m64 __a, __m64 __b)
773 {
774 return (__m64)__builtin_ia32_pavgw((__v4hi)__a, (__v4hi)__b);
775 }
776
777 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_sad_pu8(__m64 __a,__m64 __b)778 _mm_sad_pu8(__m64 __a, __m64 __b)
779 {
780 return (__m64)__builtin_ia32_psadbw((__v8qi)__a, (__v8qi)__b);
781 }
782
783 static __inline__ unsigned int __DEFAULT_FN_ATTRS
_mm_getcsr(void)784 _mm_getcsr(void)
785 {
786 return __builtin_ia32_stmxcsr();
787 }
788
789 static __inline__ void __DEFAULT_FN_ATTRS
_mm_setcsr(unsigned int __i)790 _mm_setcsr(unsigned int __i)
791 {
792 __builtin_ia32_ldmxcsr(__i);
793 }
794
795 #define _mm_shuffle_ps(a, b, mask) __extension__ ({ \
796 (__m128)__builtin_shufflevector((__v4sf)(__m128)(a), (__v4sf)(__m128)(b), \
797 (mask) & 0x3, ((mask) & 0xc) >> 2, \
798 (((mask) & 0x30) >> 4) + 4, \
799 (((mask) & 0xc0) >> 6) + 4); })
800
801 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_unpackhi_ps(__m128 __a,__m128 __b)802 _mm_unpackhi_ps(__m128 __a, __m128 __b)
803 {
804 return __builtin_shufflevector(__a, __b, 2, 6, 3, 7);
805 }
806
807 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_unpacklo_ps(__m128 __a,__m128 __b)808 _mm_unpacklo_ps(__m128 __a, __m128 __b)
809 {
810 return __builtin_shufflevector(__a, __b, 0, 4, 1, 5);
811 }
812
813 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_move_ss(__m128 __a,__m128 __b)814 _mm_move_ss(__m128 __a, __m128 __b)
815 {
816 return __builtin_shufflevector(__a, __b, 4, 1, 2, 3);
817 }
818
819 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_movehl_ps(__m128 __a,__m128 __b)820 _mm_movehl_ps(__m128 __a, __m128 __b)
821 {
822 return __builtin_shufflevector(__a, __b, 6, 7, 2, 3);
823 }
824
825 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_movelh_ps(__m128 __a,__m128 __b)826 _mm_movelh_ps(__m128 __a, __m128 __b)
827 {
828 return __builtin_shufflevector(__a, __b, 0, 1, 4, 5);
829 }
830
831 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cvtpi16_ps(__m64 __a)832 _mm_cvtpi16_ps(__m64 __a)
833 {
834 __m64 __b, __c;
835 __m128 __r;
836
837 __b = _mm_setzero_si64();
838 __b = _mm_cmpgt_pi16(__b, __a);
839 __c = _mm_unpackhi_pi16(__a, __b);
840 __r = _mm_setzero_ps();
841 __r = _mm_cvtpi32_ps(__r, __c);
842 __r = _mm_movelh_ps(__r, __r);
843 __c = _mm_unpacklo_pi16(__a, __b);
844 __r = _mm_cvtpi32_ps(__r, __c);
845
846 return __r;
847 }
848
849 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cvtpu16_ps(__m64 __a)850 _mm_cvtpu16_ps(__m64 __a)
851 {
852 __m64 __b, __c;
853 __m128 __r;
854
855 __b = _mm_setzero_si64();
856 __c = _mm_unpackhi_pi16(__a, __b);
857 __r = _mm_setzero_ps();
858 __r = _mm_cvtpi32_ps(__r, __c);
859 __r = _mm_movelh_ps(__r, __r);
860 __c = _mm_unpacklo_pi16(__a, __b);
861 __r = _mm_cvtpi32_ps(__r, __c);
862
863 return __r;
864 }
865
866 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cvtpi8_ps(__m64 __a)867 _mm_cvtpi8_ps(__m64 __a)
868 {
869 __m64 __b;
870
871 __b = _mm_setzero_si64();
872 __b = _mm_cmpgt_pi8(__b, __a);
873 __b = _mm_unpacklo_pi8(__a, __b);
874
875 return _mm_cvtpi16_ps(__b);
876 }
877
878 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cvtpu8_ps(__m64 __a)879 _mm_cvtpu8_ps(__m64 __a)
880 {
881 __m64 __b;
882
883 __b = _mm_setzero_si64();
884 __b = _mm_unpacklo_pi8(__a, __b);
885
886 return _mm_cvtpi16_ps(__b);
887 }
888
889 static __inline__ __m128 __DEFAULT_FN_ATTRS
_mm_cvtpi32x2_ps(__m64 __a,__m64 __b)890 _mm_cvtpi32x2_ps(__m64 __a, __m64 __b)
891 {
892 __m128 __c;
893
894 __c = _mm_setzero_ps();
895 __c = _mm_cvtpi32_ps(__c, __b);
896 __c = _mm_movelh_ps(__c, __c);
897
898 return _mm_cvtpi32_ps(__c, __a);
899 }
900
901 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_cvtps_pi16(__m128 __a)902 _mm_cvtps_pi16(__m128 __a)
903 {
904 __m64 __b, __c;
905
906 __b = _mm_cvtps_pi32(__a);
907 __a = _mm_movehl_ps(__a, __a);
908 __c = _mm_cvtps_pi32(__a);
909
910 return _mm_packs_pi32(__b, __c);
911 }
912
913 static __inline__ __m64 __DEFAULT_FN_ATTRS
_mm_cvtps_pi8(__m128 __a)914 _mm_cvtps_pi8(__m128 __a)
915 {
916 __m64 __b, __c;
917
918 __b = _mm_cvtps_pi16(__a);
919 __c = _mm_setzero_si64();
920
921 return _mm_packs_pi16(__b, __c);
922 }
923
924 static __inline__ int __DEFAULT_FN_ATTRS
_mm_movemask_ps(__m128 __a)925 _mm_movemask_ps(__m128 __a)
926 {
927 return __builtin_ia32_movmskps(__a);
928 }
929
930
931 #ifdef _MSC_VER
932 #define _MM_ALIGN16 __declspec(align(16))
933 #endif
934
935 #define _MM_SHUFFLE(z, y, x, w) (((z) << 6) | ((y) << 4) | ((x) << 2) | (w))
936
937 #define _MM_EXCEPT_INVALID (0x0001)
938 #define _MM_EXCEPT_DENORM (0x0002)
939 #define _MM_EXCEPT_DIV_ZERO (0x0004)
940 #define _MM_EXCEPT_OVERFLOW (0x0008)
941 #define _MM_EXCEPT_UNDERFLOW (0x0010)
942 #define _MM_EXCEPT_INEXACT (0x0020)
943 #define _MM_EXCEPT_MASK (0x003f)
944
945 #define _MM_MASK_INVALID (0x0080)
946 #define _MM_MASK_DENORM (0x0100)
947 #define _MM_MASK_DIV_ZERO (0x0200)
948 #define _MM_MASK_OVERFLOW (0x0400)
949 #define _MM_MASK_UNDERFLOW (0x0800)
950 #define _MM_MASK_INEXACT (0x1000)
951 #define _MM_MASK_MASK (0x1f80)
952
953 #define _MM_ROUND_NEAREST (0x0000)
954 #define _MM_ROUND_DOWN (0x2000)
955 #define _MM_ROUND_UP (0x4000)
956 #define _MM_ROUND_TOWARD_ZERO (0x6000)
957 #define _MM_ROUND_MASK (0x6000)
958
959 #define _MM_FLUSH_ZERO_MASK (0x8000)
960 #define _MM_FLUSH_ZERO_ON (0x8000)
961 #define _MM_FLUSH_ZERO_OFF (0x0000)
962
963 #define _MM_GET_EXCEPTION_MASK() (_mm_getcsr() & _MM_MASK_MASK)
964 #define _MM_GET_EXCEPTION_STATE() (_mm_getcsr() & _MM_EXCEPT_MASK)
965 #define _MM_GET_FLUSH_ZERO_MODE() (_mm_getcsr() & _MM_FLUSH_ZERO_MASK)
966 #define _MM_GET_ROUNDING_MODE() (_mm_getcsr() & _MM_ROUND_MASK)
967
968 #define _MM_SET_EXCEPTION_MASK(x) (_mm_setcsr((_mm_getcsr() & ~_MM_MASK_MASK) | (x)))
969 #define _MM_SET_EXCEPTION_STATE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_EXCEPT_MASK) | (x)))
970 #define _MM_SET_FLUSH_ZERO_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_FLUSH_ZERO_MASK) | (x)))
971 #define _MM_SET_ROUNDING_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_ROUND_MASK) | (x)))
972
973 #define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) \
974 do { \
975 __m128 tmp3, tmp2, tmp1, tmp0; \
976 tmp0 = _mm_unpacklo_ps((row0), (row1)); \
977 tmp2 = _mm_unpacklo_ps((row2), (row3)); \
978 tmp1 = _mm_unpackhi_ps((row0), (row1)); \
979 tmp3 = _mm_unpackhi_ps((row2), (row3)); \
980 (row0) = _mm_movelh_ps(tmp0, tmp2); \
981 (row1) = _mm_movehl_ps(tmp2, tmp0); \
982 (row2) = _mm_movelh_ps(tmp1, tmp3); \
983 (row3) = _mm_movehl_ps(tmp3, tmp1); \
984 } while (0)
985
986 /* Aliases for compatibility. */
987 #define _m_pextrw _mm_extract_pi16
988 #define _m_pinsrw _mm_insert_pi16
989 #define _m_pmaxsw _mm_max_pi16
990 #define _m_pmaxub _mm_max_pu8
991 #define _m_pminsw _mm_min_pi16
992 #define _m_pminub _mm_min_pu8
993 #define _m_pmovmskb _mm_movemask_pi8
994 #define _m_pmulhuw _mm_mulhi_pu16
995 #define _m_pshufw _mm_shuffle_pi16
996 #define _m_maskmovq _mm_maskmove_si64
997 #define _m_pavgb _mm_avg_pu8
998 #define _m_pavgw _mm_avg_pu16
999 #define _m_psadbw _mm_sad_pu8
1000 #define _m_ _mm_
1001 #define _m_ _mm_
1002
1003 #undef __DEFAULT_FN_ATTRS
1004
1005 /* Ugly hack for backwards-compatibility (compatible with gcc) */
1006 #if defined(__SSE2__) && !__has_feature(modules)
1007 #include <emmintrin.h>
1008 #endif
1009
1010 #endif /* __XMMINTRIN_H */
1011